diff --git a/recipes/Cluster_create_RayCluster.md b/recipes/Cluster_create_RayCluster.md
index 6ce311fb6..ea9be8e25 100644
--- a/recipes/Cluster_create_RayCluster.md
+++ b/recipes/Cluster_create_RayCluster.md
@@ -10,8 +10,8 @@ $ xpk cluster create-ray --project=golden-project --zone=us-central1-a --cluster
 [XPK] Starting xpk v0.0.0
 [XPK] Starting cluster create for cluster golden-cluster:
 [XPK] Working on golden-project and us-central1-a
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a
+[XPK] Task: `Get reservation golden-reservation` is implemented by the following command not running since it is a dry run. 
+gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a --format="json(specificReservation,aggregateReservation,status,deploymentType,resourcePolicies)"
 [XPK] Task: `Determine server supported GKE versions for default gke version` is implemented by the following command not running since it is a dry run. 
 gcloud container get-server-config --project=golden-project --region=us-central1 --flatten="channels" --filter="channels.channel=RAPID" --format="value(channels.defaultVersion)"
 [XPK] Task: `Determine server supported GKE versions for valid versions` is implemented by the following command not running since it is a dry run. 
@@ -50,8 +50,6 @@ gcloud beta container clusters describe golden-cluster --location us-central1 --
 We assume that the underlying system is: SystemCharacteristics(topology='2x2x1', vms_per_slice=1, gke_accelerator='tpu7x', gce_machine_type='tpu7x-standard-4t', chips_per_vm=4, accelerator_type=TPU, device_type='tpu7x-8', supports_sub_slicing=False, supports_super_slicing=False, supports_accelerator_network_profile=False, docker_platform=<DockerPlatform.AMD: 'linux/amd64'>, requires_workload_policy=False, gpu_config=None, parallel_containers=2)
 [XPK] Task: `Get All Node Pools` is implemented by the following command not running since it is a dry run. 
 gcloud beta container node-pools list --cluster golden-cluster --project=golden-project --location=us-central1 --format="csv[no-heading](name)"
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a
 [XPK] Creating 1 node pool or pools of tpu7x-8
 Underlyingly, we assume that means: SystemCharacteristics(topology='2x2x1', vms_per_slice=1, gke_accelerator='tpu7x', gce_machine_type='tpu7x-standard-4t', chips_per_vm=4, accelerator_type=TPU, device_type='tpu7x-8', supports_sub_slicing=False, supports_super_slicing=False, supports_accelerator_network_profile=False, docker_platform=<DockerPlatform.AMD: 'linux/amd64'>, requires_workload_policy=False, gpu_config=None, parallel_containers=2)
 [XPK] Task: `Get Node Pool Zone` is implemented by the following command not running since it is a dry run. 
@@ -64,8 +62,6 @@ kubectl get configmap golden-cluster-resources-configmap -o=custom-columns="Conf
 [XPK] Pretending all the jobs succeeded
 [XPK] Create or delete node pool request complete.
 [XPK] Creating ConfigMap for cluster
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a
 [XPK] Temp file (0604d72ef175c94fc796d8f02cff009b4241e85d444d22d414a56a47764d7bbb) content: 
 kind: ConfigMap
 apiVersion: v1
diff --git a/recipes/Cluster_create_private.md b/recipes/Cluster_create_private.md
index 3c67d9539..f7b2341b8 100644
--- a/recipes/Cluster_create_private.md
+++ b/recipes/Cluster_create_private.md
@@ -12,8 +12,8 @@ $ xpk cluster create-pathways --project=golden-project --zone=us-central1-a --cl
 [XPK] Working on golden-project and us-central1-a
 [XPK] Task: `Retrieve available pathways machine types` is implemented by the following command not running since it is a dry run. 
 gcloud compute machine-types list --filter "guestCpus >= 49 AND memoryMb >= 238592 AND zone = 'us-central1-a'" --format="value(name)" --project=golden-project
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a
+[XPK] Task: `Get reservation golden-reservation` is implemented by the following command not running since it is a dry run. 
+gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a --format="json(specificReservation,aggregateReservation,status,deploymentType,resourcePolicies)"
 [XPK] Task: `Determine server supported GKE versions for default gke version` is implemented by the following command not running since it is a dry run. 
 gcloud container get-server-config --project=golden-project --region=us-central1 --flatten="channels" --filter="channels.channel=RAPID" --format="value(channels.defaultVersion)"
 [XPK] Task: `Determine server supported GKE versions for valid versions` is implemented by the following command not running since it is a dry run. 
@@ -54,8 +54,6 @@ gcloud beta container clusters describe golden-cluster-private --location us-cen
 We assume that the underlying system is: SystemCharacteristics(topology='2x2x1', vms_per_slice=1, gke_accelerator='tpu-v5p-slice', gce_machine_type='ct5p-hightpu-4t', chips_per_vm=4, accelerator_type=TPU, device_type='v5p-8', supports_sub_slicing=False, supports_super_slicing=False, supports_accelerator_network_profile=False, docker_platform=<DockerPlatform.AMD: 'linux/amd64'>, requires_workload_policy=False, gpu_config=None, parallel_containers=1)
 [XPK] Task: `Get All Node Pools` is implemented by the following command not running since it is a dry run. 
 gcloud beta container node-pools list --cluster golden-cluster-private --project=golden-project --location=us-central1 --format="csv[no-heading](name)"
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a
 [XPK] Creating 1 node pool or pools of v5p-8
 Underlyingly, we assume that means: SystemCharacteristics(topology='2x2x1', vms_per_slice=1, gke_accelerator='tpu-v5p-slice', gce_machine_type='ct5p-hightpu-4t', chips_per_vm=4, accelerator_type=TPU, device_type='v5p-8', supports_sub_slicing=False, supports_super_slicing=False, supports_accelerator_network_profile=False, docker_platform=<DockerPlatform.AMD: 'linux/amd64'>, requires_workload_policy=False, gpu_config=None, parallel_containers=1)
 [XPK] Task: `Get Node Pool Zone` is implemented by the following command not running since it is a dry run. 
@@ -69,8 +67,6 @@ kubectl get configmap golden-cluster-private-resources-configmap -o=custom-colum
 [XPK] Pretending all the jobs succeeded
 [XPK] Create or delete node pool request complete.
 [XPK] Creating ConfigMap for cluster
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a
 [XPK] Temp file (8669497cfbe494756d36922054f924d7dca463141f0e5d0329e517c880cf2f06) content: 
 kind: ConfigMap
 apiVersion: v1
diff --git a/recipes/Cluster_create_sub-slicing.md b/recipes/Cluster_create_sub-slicing.md
index b03f037e1..ea1e6c2a1 100644
--- a/recipes/Cluster_create_sub-slicing.md
+++ b/recipes/Cluster_create_sub-slicing.md
@@ -10,10 +10,8 @@ $ SUB_SLICING_ENABLED=true xpk cluster create --project=golden-project --zone=us
 [XPK] Starting xpk v0.0.0
 [XPK] Starting cluster create for cluster golden-cluster:
 [XPK] Working on golden-project and us-central1-a
-[XPK] Task: `Get reservation deployment type` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a --format="value(deploymentType)"
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a
+[XPK] Task: `Get reservation golden-reservation` is implemented by the following command not running since it is a dry run. 
+gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a --format="json(specificReservation,aggregateReservation,status,deploymentType,resourcePolicies)"
 [XPK] Task: `Determine server supported GKE versions for default gke version` is implemented by the following command not running since it is a dry run. 
 gcloud container get-server-config --project=golden-project --region=us-central1 --flatten="channels" --filter="channels.channel=RAPID" --format="value(channels.defaultVersion)"
 [XPK] Task: `Determine server supported GKE versions for valid versions` is implemented by the following command not running since it is a dry run. 
@@ -52,8 +50,6 @@ gcloud beta container clusters describe golden-cluster --location us-central1 --
 We assume that the underlying system is: SystemCharacteristics(topology='4x4', vms_per_slice=4, gke_accelerator='tpu-v6e-slice', gce_machine_type='ct6e-standard-4t', chips_per_vm=4, accelerator_type=TPU, device_type='v6e-16', supports_sub_slicing=True, supports_super_slicing=False, supports_accelerator_network_profile=True, docker_platform=<DockerPlatform.AMD: 'linux/amd64'>, requires_workload_policy=False, gpu_config=None, parallel_containers=1)
 [XPK] Task: `Get All Node Pools` is implemented by the following command not running since it is a dry run. 
 gcloud beta container node-pools list --cluster golden-cluster --project=golden-project --location=us-central1 --format="csv[no-heading](name)"
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a
 [XPK] Creating 1 node pool or pools of v6e-16
 Underlyingly, we assume that means: SystemCharacteristics(topology='4x4', vms_per_slice=4, gke_accelerator='tpu-v6e-slice', gce_machine_type='ct6e-standard-4t', chips_per_vm=4, accelerator_type=TPU, device_type='v6e-16', supports_sub_slicing=True, supports_super_slicing=False, supports_accelerator_network_profile=True, docker_platform=<DockerPlatform.AMD: 'linux/amd64'>, requires_workload_policy=False, gpu_config=None, parallel_containers=1)
 [XPK] Task: `Get Node Pool Zone` is implemented by the following command not running since it is a dry run. 
@@ -66,8 +62,6 @@ kubectl get configmap golden-cluster-resources-configmap -o=custom-columns="Conf
 [XPK] Pretending all the jobs succeeded
 [XPK] Create or delete node pool request complete.
 [XPK] Creating ConfigMap for cluster
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a
 [XPK] Temp file (8d0f4b1e96d79a5d572cbb1a403ac3285b6a9390b6092b86a76bf66705e35d44) content: 
 kind: ConfigMap
 apiVersion: v1
diff --git a/recipes/Cluster_create_super-slicing.md b/recipes/Cluster_create_super-slicing.md
index 28ac9f6cf..df0f21252 100644
--- a/recipes/Cluster_create_super-slicing.md
+++ b/recipes/Cluster_create_super-slicing.md
@@ -1,19 +1,19 @@
 # Cluster create super-slicing
+
 Creates a GKE cluster with TPU super-slicing enabled for multi-slice training.
 
 # Running the command
+
 ```shell #golden
-xpk cluster create --project=golden-project --zone=us-central1-a --cluster=golden-cluster --tpu-type=tpu7x-4x4x4 --reservation=golden-reservation/reservationBlocks/block/reservationSubBlocks/subblock --super-slicing --num-cubes=5
+DRY_RUN_RESERVATION_SUB_BLOCKS='[{"name": "sub0", "count": 16, "inUseCount": 0}, {"name": "sub1", "count": 16, "inUseCount": 0}, {"name": "sub2", "count": 16, "inUseCount": 15}, {"name": "sub3", "count": 16, "inUseCount": 0}]' xpk cluster create --project=golden-project --zone=us-central1-a --cluster=golden-cluster --tpu-type=tpu7x-4x4x4 --reservation=golden-reservation/reservationBlocks/block --super-slicing --num-cubes=3
 ```
 <!--
-$ xpk cluster create --project=golden-project --zone=us-central1-a --cluster=golden-cluster --tpu-type=tpu7x-4x4x4 --reservation=golden-reservation/reservationBlocks/block/reservationSubBlocks/subblock --super-slicing --num-cubes=5
+$ DRY_RUN_RESERVATION_SUB_BLOCKS='[{"name": "sub0", "count": 16, "inUseCount": 0}, {"name": "sub1", "count": 16, "inUseCount": 0}, {"name": "sub2", "count": 16, "inUseCount": 15}, {"name": "sub3", "count": 16, "inUseCount": 0}]' xpk cluster create --project=golden-project --zone=us-central1-a --cluster=golden-cluster --tpu-type=tpu7x-4x4x4 --reservation=golden-reservation/reservationBlocks/block --super-slicing --num-cubes=3
 [XPK] Starting xpk v0.0.0
 [XPK] Starting cluster create for cluster golden-cluster:
 [XPK] Working on golden-project and us-central1-a
-[XPK] Task: `Get reservation deployment type` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a --format="value(deploymentType)"
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a
+[XPK] Task: `Get reservation golden-reservation` is implemented by the following command not running since it is a dry run. 
+gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a --format="json(specificReservation,aggregateReservation,status,deploymentType,resourcePolicies)"
 [XPK] Task: `Determine server supported GKE versions for default gke version` is implemented by the following command not running since it is a dry run. 
 gcloud container get-server-config --project=golden-project --region=us-central1 --flatten="channels" --filter="channels.channel=RAPID" --format="value(channels.defaultVersion)"
 [XPK] Task: `Determine server supported GKE versions for valid versions` is implemented by the following command not running since it is a dry run. 
@@ -48,13 +48,11 @@ kubectl wait deployment/coredns --for=condition=Available=true --namespace=kube-
 [XPK] Skipping CoreDNS deployment since it already exists.
 [XPK] Task: `Determine current gke master version` is implemented by the following command not running since it is a dry run. 
 gcloud beta container clusters describe golden-cluster --location us-central1 --project golden-project --format="value(currentMasterVersion)"
-[XPK] Creating 5 node pool or pools of tpu7x-4x4x4
+[XPK] Creating 3 node pool or pools of tpu7x-4x4x4
 We assume that the underlying system is: SystemCharacteristics(topology='4x4x4', vms_per_slice=16, gke_accelerator='tpu7x', gce_machine_type='tpu7x-standard-4t', chips_per_vm=4, accelerator_type=TPU, device_type='tpu7x-128', supports_sub_slicing=False, supports_super_slicing=True, supports_accelerator_network_profile=False, docker_platform=<DockerPlatform.AMD: 'linux/amd64'>, requires_workload_policy=True, gpu_config=None, parallel_containers=2)
 [XPK] Task: `Get All Node Pools` is implemented by the following command not running since it is a dry run. 
 gcloud beta container node-pools list --cluster golden-cluster --project=golden-project --location=us-central1 --format="csv[no-heading](name)"
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a
-[XPK] Creating 5 node pool or pools of tpu7x-128
+[XPK] Creating 3 node pool or pools of tpu7x-128
 Underlyingly, we assume that means: SystemCharacteristics(topology='4x4x4', vms_per_slice=16, gke_accelerator='tpu7x', gce_machine_type='tpu7x-standard-4t', chips_per_vm=4, accelerator_type=TPU, device_type='tpu7x-128', supports_sub_slicing=False, supports_super_slicing=True, supports_accelerator_network_profile=False, docker_platform=<DockerPlatform.AMD: 'linux/amd64'>, requires_workload_policy=True, gpu_config=None, parallel_containers=2)
 [XPK] Task: `Get Node Pool Zone` is implemented by the following command not running since it is a dry run. 
 gcloud beta container node-pools describe 0 --cluster golden-cluster --project=golden-project --location=us-central1 --format="value(locations)"
@@ -63,26 +61,24 @@ kubectl get configmap golden-cluster-resources-configmap -o=custom-columns="Conf
 [XPK] Existing node pool names  ['0']
 [XPK] Task: `Retrieve resource policy` is implemented by the following command not running since it is a dry run. 
 gcloud beta compute resource-policies describe tpu7x-128-4x4x4-ss-placement-policy --project=golden-project --region=us-central1
-[XPK] To complete NodepoolCreate-golden-cluster-np-0 we are executing gcloud beta container node-pools create golden-cluster-np-0 --location=us-central1 --cluster=golden-cluster --project=golden-project --node-locations=us-central1-a --machine-type=tpu7x-standard-4t --host-maintenance-interval=AS_NEEDED --reservation-affinity=specific --reservation=golden-reservation/reservationBlocks/block/reservationSubBlocks/subblock --placement-policy=tpu7x-128-4x4x4-ss-placement-policy --enable-gvnic --node-version=0 --num-nodes=16 --scopes=storage-full,gke-default,"https://www.googleapis.com/auth/cloud-platform" --max-pods-per-node 15  
-[XPK] To complete NodepoolCreate-golden-cluster-np-1 we are executing gcloud beta container node-pools create golden-cluster-np-1 --location=us-central1 --cluster=golden-cluster --project=golden-project --node-locations=us-central1-a --machine-type=tpu7x-standard-4t --host-maintenance-interval=AS_NEEDED --reservation-affinity=specific --reservation=golden-reservation/reservationBlocks/block/reservationSubBlocks/subblock --placement-policy=tpu7x-128-4x4x4-ss-placement-policy --enable-gvnic --node-version=0 --num-nodes=16 --scopes=storage-full,gke-default,"https://www.googleapis.com/auth/cloud-platform" --max-pods-per-node 15  
-[XPK] To complete NodepoolCreate-golden-cluster-np-2 we are executing gcloud beta container node-pools create golden-cluster-np-2 --location=us-central1 --cluster=golden-cluster --project=golden-project --node-locations=us-central1-a --machine-type=tpu7x-standard-4t --host-maintenance-interval=AS_NEEDED --reservation-affinity=specific --reservation=golden-reservation/reservationBlocks/block/reservationSubBlocks/subblock --placement-policy=tpu7x-128-4x4x4-ss-placement-policy --enable-gvnic --node-version=0 --num-nodes=16 --scopes=storage-full,gke-default,"https://www.googleapis.com/auth/cloud-platform" --max-pods-per-node 15  
-[XPK] To complete NodepoolCreate-golden-cluster-np-3 we are executing gcloud beta container node-pools create golden-cluster-np-3 --location=us-central1 --cluster=golden-cluster --project=golden-project --node-locations=us-central1-a --machine-type=tpu7x-standard-4t --host-maintenance-interval=AS_NEEDED --reservation-affinity=specific --reservation=golden-reservation/reservationBlocks/block/reservationSubBlocks/subblock --placement-policy=tpu7x-128-4x4x4-ss-placement-policy --enable-gvnic --node-version=0 --num-nodes=16 --scopes=storage-full,gke-default,"https://www.googleapis.com/auth/cloud-platform" --max-pods-per-node 15  
-[XPK] To complete NodepoolCreate-golden-cluster-np-4 we are executing gcloud beta container node-pools create golden-cluster-np-4 --location=us-central1 --cluster=golden-cluster --project=golden-project --node-locations=us-central1-a --machine-type=tpu7x-standard-4t --host-maintenance-interval=AS_NEEDED --reservation-affinity=specific --reservation=golden-reservation/reservationBlocks/block/reservationSubBlocks/subblock --placement-policy=tpu7x-128-4x4x4-ss-placement-policy --enable-gvnic --node-version=0 --num-nodes=16 --scopes=storage-full,gke-default,"https://www.googleapis.com/auth/cloud-platform" --max-pods-per-node 15  
-[XPK] Breaking up a total of 5 commands into 1 batches
+[XPK] Task: `Count healthy fitting sub-blocks in block` is implemented by the following command not running since it is a dry run. 
+gcloud beta compute reservations sub-blocks list golden-reservation --block-name=block --project=golden-project --zone=us-central1-a --filter="healthInfo.healthStatus=HEALTHY" --format="json(name,count,inUseCount)"
+[XPK] To complete NodepoolCreate-golden-cluster-np-0 we are executing gcloud beta container node-pools create golden-cluster-np-0 --location=us-central1 --cluster=golden-cluster --project=golden-project --node-locations=us-central1-a --machine-type=tpu7x-standard-4t --host-maintenance-interval=AS_NEEDED --reservation-affinity=specific --reservation=golden-reservation/reservationBlocks/block/reservationSubBlocks/sub0 --placement-policy=tpu7x-128-4x4x4-ss-placement-policy --enable-gvnic --node-version=0 --num-nodes=16 --scopes=storage-full,gke-default,"https://www.googleapis.com/auth/cloud-platform" --max-pods-per-node 15  
+[XPK] To complete NodepoolCreate-golden-cluster-np-1 we are executing gcloud beta container node-pools create golden-cluster-np-1 --location=us-central1 --cluster=golden-cluster --project=golden-project --node-locations=us-central1-a --machine-type=tpu7x-standard-4t --host-maintenance-interval=AS_NEEDED --reservation-affinity=specific --reservation=golden-reservation/reservationBlocks/block/reservationSubBlocks/sub1 --placement-policy=tpu7x-128-4x4x4-ss-placement-policy --enable-gvnic --node-version=0 --num-nodes=16 --scopes=storage-full,gke-default,"https://www.googleapis.com/auth/cloud-platform" --max-pods-per-node 15  
+[XPK] To complete NodepoolCreate-golden-cluster-np-2 we are executing gcloud beta container node-pools create golden-cluster-np-2 --location=us-central1 --cluster=golden-cluster --project=golden-project --node-locations=us-central1-a --machine-type=tpu7x-standard-4t --host-maintenance-interval=AS_NEEDED --reservation-affinity=specific --reservation=golden-reservation/reservationBlocks/block/reservationSubBlocks/sub3 --placement-policy=tpu7x-128-4x4x4-ss-placement-policy --enable-gvnic --node-version=0 --num-nodes=16 --scopes=storage-full,gke-default,"https://www.googleapis.com/auth/cloud-platform" --max-pods-per-node 15  
+[XPK] Breaking up a total of 3 commands into 1 batches
 [XPK] Pretending all the jobs succeeded
 [XPK] Create or delete node pool request complete.
 [XPK] Creating ConfigMap for cluster
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a
-[XPK] Temp file (fdab178a48d8bf35a601075f72c230133f9e6f665c6bd45129ed0185d5b00aec) content: 
+[XPK] Temp file (92ab8eba56c8168041408f03de0995c86cd9bf277397f4db955aef042b0b26cd) content: 
 kind: ConfigMap
 apiVersion: v1
 metadata:
   name: golden-cluster-resources-configmap
 data:
-  tpu7x-128: "80"
+  tpu7x-128: "48"
 
-[XPK] Temp file (cd1d35a150fd66bdee3f462567a1cf4ed56b8af9ab774e15dcdac83b6b82e48e) content: 
+[XPK] Temp file (d6354673799dbf5d6ac92c682900bc8185116e90580c3dff5355cb69dc2b2a7e) content: 
 kind: ConfigMap
 apiVersion: v1
 metadata:
@@ -90,7 +86,7 @@ metadata:
 data:
   xpk_version: v0.0.0
   capacity_type: RESERVATION
-  reservation_id: golden-reservation/reservationBlocks/block/reservationSubBlocks/subblock
+  reservation_id: golden-reservation/reservationBlocks/block
 
 [XPK] Breaking up a total of 2 commands into 1 batches
 [XPK] Pretending all the jobs succeeded
@@ -200,12 +196,12 @@ kubectl get deployment kueue-controller-manager -n kueue-system -o jsonpath='{.s
 kubectl apply --server-side --force-conflicts -f https://github.com/kubernetes-sigs/kueue/releases/download/v0.15.2/manifests.yaml
 [XPK] Task: `Wait for Kueue to be available` is implemented by the following command not running since it is a dry run. 
 kubectl wait deploy/kueue-controller-manager -n kueue-system --for=condition=available --timeout=10m
-[XPK] Temp file (6df31e8df3d8970d7ed3bf3aa948ae7cea9487c15ed6cfb1577ca6c948cf5525) content: 
+[XPK] Temp file (20ee412fd0eeeaf2c32856f66404b54a7a638654ffbf9afb7d8f50780311ed10) content: 
 
 apiVersion: kueue.x-k8s.io/v1beta1
 kind: ResourceFlavor
 metadata:
-  name: "5xtpu7x-128"
+  name: "3xtpu7x-128"
 spec:
   nodeLabels: {"cloud.google.com/gke-tpu-accelerator": "tpu7x"}
   topologyName: super-slice-topology
@@ -239,7 +235,7 @@ spec:
     reclaimWithinCohort: Never # Don't preempt other queues in the cohort.
     withinClusterQueue: LowerPriority
   namespaceSelector: {} # match all.
-  resourceGroups: [{'coveredResources': ['google.com/tpu'], 'flavors': [{'name': '5xtpu7x-128', 'resources': [{'name': 'google.com/tpu', 'nominalQuota': 320}]}]}]
+  resourceGroups: [{'coveredResources': ['google.com/tpu'], 'flavors': [{'name': '3xtpu7x-128', 'resources': [{'name': 'google.com/tpu', 'nominalQuota': 192}]}]}]
   admissionChecks:
   - ss-kueue-operator
 ---
@@ -301,7 +297,7 @@ spec:
   - nodeLabel: cloud.google.com/gke-tpu-partition-4x4x4-id
   - nodeLabel: kubernetes.io/hostname
 [XPK] Task: `Applying Kueue Custom Resources` is implemented by the following command not running since it is a dry run. 
-kubectl apply -f 6df31e8df3d8970d7ed3bf3aa948ae7cea9487c15ed6cfb1577ca6c948cf5525
+kubectl apply -f 20ee412fd0eeeaf2c32856f66404b54a7a638654ffbf9afb7d8f50780311ed10
 [XPK] Try 1: Updating Controller Manager resources
 [XPK] Task: `Updating Controller Manager resources` is implemented by the following command not running since it is a dry run. 
 kubectl patch deployment kueue-controller-manager -n kueue-system --type='strategic' --patch='{"spec": {"replicas": 3, "template": {"spec": {"containers": [{"name": "manager", "resources": {"requests": {"cpu": "16", "memory": "64Gi"}, "limits": {"cpu": "16", "memory": "64Gi"}}}]}}}}'
diff --git a/recipes/Cluster_create_with_gb200-4.md b/recipes/Cluster_create_with_gb200-4.md
index f7aeead70..1ce59ce44 100644
--- a/recipes/Cluster_create_with_gb200-4.md
+++ b/recipes/Cluster_create_with_gb200-4.md
@@ -10,8 +10,8 @@ $ xpk cluster create --project=golden-project --zone=us-central1-a --cluster=gol
 [XPK] Starting xpk v0.0.0
 [XPK] Starting cluster create for cluster golden-cluster:
 [XPK] Working on golden-project and us-central1-a
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a
+[XPK] Task: `Get reservation golden-reservation` is implemented by the following command not running since it is a dry run. 
+gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a --format="json(specificReservation,aggregateReservation,status,deploymentType,resourcePolicies)"
 [XPK] Task: `Determine server supported GKE versions for default gke version` is implemented by the following command not running since it is a dry run. 
 gcloud container get-server-config --project=golden-project --region=us-central1 --flatten="channels" --filter="channels.channel=RAPID" --format="value(channels.defaultVersion)"
 [XPK] Task: `Determine server supported GKE versions for valid versions` is implemented by the following command not running since it is a dry run. 
@@ -50,8 +50,6 @@ gcloud beta container clusters describe golden-cluster --location us-central1 --
 We assume that the underlying system is: SystemCharacteristics(topology='1x72', vms_per_slice=1, gke_accelerator='nvidia-gb200', gce_machine_type='a4x-highgpu-4g', chips_per_vm=4, accelerator_type=GPU, device_type='gb200-4', supports_sub_slicing=False, supports_super_slicing=False, supports_accelerator_network_profile=True, docker_platform=<DockerPlatform.ARM: 'linux/arm64'>, requires_workload_policy=True, gpu_config=GpuConfig(requires_topology=True, gpu_direct_name='rdma', nccl_installer='https://raw.githubusercontent.com/GoogleCloudPlatform/container-engine-accelerators/master/gpudirect-rdma/nccl-rdma-installer-a4x.yaml', jobset_decorator_fn=<function decorate_jobset>), parallel_containers=1)
 [XPK] Task: `Get All Node Pools` is implemented by the following command not running since it is a dry run. 
 gcloud beta container node-pools list --cluster golden-cluster --project=golden-project --location=us-central1 --format="csv[no-heading](name)"
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a
 [XPK] Creating 1 node pool with 2 nodes of gb200-4
 Underlyingly, we assume that means: SystemCharacteristics(topology='1x72', vms_per_slice=1, gke_accelerator='nvidia-gb200', gce_machine_type='a4x-highgpu-4g', chips_per_vm=4, accelerator_type=GPU, device_type='gb200-4', supports_sub_slicing=False, supports_super_slicing=False, supports_accelerator_network_profile=True, docker_platform=<DockerPlatform.ARM: 'linux/arm64'>, requires_workload_policy=True, gpu_config=GpuConfig(requires_topology=True, gpu_direct_name='rdma', nccl_installer='https://raw.githubusercontent.com/GoogleCloudPlatform/container-engine-accelerators/master/gpudirect-rdma/nccl-rdma-installer-a4x.yaml', jobset_decorator_fn=<function decorate_jobset>), parallel_containers=1)
 [XPK] Task: `Get Node Pool Zone` is implemented by the following command not running since it is a dry run. 
@@ -66,8 +64,6 @@ gcloud beta compute resource-policies describe gb200-4-1x72-placement-policy --p
 [XPK] Pretending all the jobs succeeded
 [XPK] Create or delete node pool request complete.
 [XPK] Creating ConfigMap for cluster
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=golden-project --zone=us-central1-a
 [XPK] Temp file (9476f7fa10da99ed4e0797d6d660cda076b5d6dfcd366a9e2560681f82697e99) content: 
 kind: ConfigMap
 apiVersion: v1
diff --git a/recipes/Cluster_create_with_shared_reservation.md b/recipes/Cluster_create_with_shared_reservation.md
index 822d9356b..8d8a55b9d 100644
--- a/recipes/Cluster_create_with_shared_reservation.md
+++ b/recipes/Cluster_create_with_shared_reservation.md
@@ -10,8 +10,8 @@ $ xpk cluster create --project=golden-project --zone=us-central1-a --cluster=gol
 [XPK] Starting xpk v0.0.0
 [XPK] Starting cluster create for cluster golden-cluster:
 [XPK] Working on golden-project and us-central1-a
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=reservation-project --zone=us-central1-a
+[XPK] Task: `Get reservation golden-reservation` is implemented by the following command not running since it is a dry run. 
+gcloud beta compute reservations describe golden-reservation --project=reservation-project --zone=us-central1-a --format="json(specificReservation,aggregateReservation,status,deploymentType,resourcePolicies)"
 [XPK] Task: `Determine server supported GKE versions for default gke version` is implemented by the following command not running since it is a dry run. 
 gcloud container get-server-config --project=golden-project --region=us-central1 --flatten="channels" --filter="channels.channel=RAPID" --format="value(channels.defaultVersion)"
 [XPK] Task: `Determine server supported GKE versions for valid versions` is implemented by the following command not running since it is a dry run. 
@@ -50,8 +50,6 @@ gcloud beta container clusters describe golden-cluster --location us-central1 --
 We assume that the underlying system is: SystemCharacteristics(topology='2x2x1', vms_per_slice=1, gke_accelerator='tpu7x', gce_machine_type='tpu7x-standard-4t', chips_per_vm=4, accelerator_type=TPU, device_type='tpu7x-8', supports_sub_slicing=False, supports_super_slicing=False, supports_accelerator_network_profile=False, docker_platform=<DockerPlatform.AMD: 'linux/amd64'>, requires_workload_policy=False, gpu_config=None, parallel_containers=2)
 [XPK] Task: `Get All Node Pools` is implemented by the following command not running since it is a dry run. 
 gcloud beta container node-pools list --cluster golden-cluster --project=golden-project --location=us-central1 --format="csv[no-heading](name)"
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=reservation-project --zone=us-central1-a
 [XPK] Creating 1 node pool or pools of tpu7x-8
 Underlyingly, we assume that means: SystemCharacteristics(topology='2x2x1', vms_per_slice=1, gke_accelerator='tpu7x', gce_machine_type='tpu7x-standard-4t', chips_per_vm=4, accelerator_type=TPU, device_type='tpu7x-8', supports_sub_slicing=False, supports_super_slicing=False, supports_accelerator_network_profile=False, docker_platform=<DockerPlatform.AMD: 'linux/amd64'>, requires_workload_policy=False, gpu_config=None, parallel_containers=2)
 [XPK] Task: `Get Node Pool Zone` is implemented by the following command not running since it is a dry run. 
@@ -64,8 +62,6 @@ kubectl get configmap golden-cluster-resources-configmap -o=custom-columns="Conf
 [XPK] Pretending all the jobs succeeded
 [XPK] Create or delete node pool request complete.
 [XPK] Creating ConfigMap for cluster
-[XPK] Task: `Describe reservation` is implemented by the following command not running since it is a dry run. 
-gcloud beta compute reservations describe golden-reservation --project=reservation-project --zone=us-central1-a
 [XPK] Temp file (0604d72ef175c94fc796d8f02cff009b4241e85d444d22d414a56a47764d7bbb) content: 
 kind: ConfigMap
 apiVersion: v1
diff --git a/src/xpk/commands/cluster.py b/src/xpk/commands/cluster.py
index 5a32d6783..e587c7b8d 100644
--- a/src/xpk/commands/cluster.py
+++ b/src/xpk/commands/cluster.py
@@ -254,7 +254,9 @@ def _validate_gsc_reservation(args, creation_description: str):
     xpk_exit(1)
 
   for reservation in get_reservations_list(args):
-    deployment_type = get_reservation_deployment_type(reservation=reservation)
+    deployment_type = get_reservation_deployment_type(
+        reservation_link=reservation
+    )
     if deployment_type != 'DENSE':
       xpk_print(
           'Error: Validation failed: The specified reservation'
diff --git a/src/xpk/core/capacity.py b/src/xpk/core/capacity.py
index e9b19edca..1f37fae06 100644
--- a/src/xpk/core/capacity.py
+++ b/src/xpk/core/capacity.py
@@ -15,12 +15,18 @@
 """
 
 import enum
-from dataclasses import dataclass
+import os
+import json
+import functools
+from dataclasses import dataclass, field
+from typing import Sequence, Any
 
 from .commands import run_command_with_updates, run_command_for_value
-from .system_characteristics import AcceleratorType
+from .system_characteristics import AcceleratorType, SystemCharacteristics
+from .gcloud_context import project_id_to_project_number
 from ..utils.console import xpk_print, xpk_exit
 from ..utils.kueue import is_queued_cluster
+from ..utils.execution_context import is_dry_run
 
 AUTOPROVISIONING_CONFIG_VALUE = 'AUTOPROVISION'
 AUTOPROVISIONING_CONFIG_MINIMUM_KEY = 'minimum_chips'
@@ -61,6 +67,191 @@ class SubBlockReservationLink(BlockReservationLink):
   sub_block_name: str
 
 
+@dataclass(frozen=True)
+class ReservationCapacity:
+  reservation: ReservationLink
+  available_slices: int
+
+
+@dataclass(frozen=True)
+class _AcceleratorResource:
+  acceleratorCount: int
+  acceleratorType: str
+
+
+@dataclass(frozen=True)
+class _SpecificReservation:
+  count: int
+  inUseCount: int
+  machine_type: str
+  guest_accelerators: list[_AcceleratorResource] = field(default_factory=list)
+  maintenance_interval: str = ''
+
+
+@dataclass(frozen=True)
+class _AggregateReservation:
+  reservedResources: list[_AcceleratorResource]
+  inUseResources: list[_AcceleratorResource]
+
+
+@dataclass(frozen=True)
+class _ReservationSubBlock:
+  name: str
+  count: int
+  in_use_count: int
+
+
+@dataclass(frozen=True)
+class _Reservation:
+  name: str
+  specificReservation: _SpecificReservation | None
+  aggregateReservation: _AggregateReservation | None
+  deployment_type: str = ''
+  resource_policy: str = ''
+
+
+def _parse_specific_reservation(data: dict[str, Any]) -> _SpecificReservation:
+  instance_properties = data.get('instanceProperties', {})
+  machine_type = instance_properties.get('machineType', '')
+  guest_accelerators_data = instance_properties.get('guestAccelerators', [])
+  guest_accelerators = [
+      _parse_accelerator_resource(acc) for acc in guest_accelerators_data
+  ]
+  maintenance_interval = instance_properties.get('maintenanceInterval', '')
+
+  return _SpecificReservation(
+      count=int(data.get('count', 0)),
+      inUseCount=int(data.get('inUseCount', 0)),
+      machine_type=machine_type,
+      guest_accelerators=guest_accelerators,
+      maintenance_interval=maintenance_interval,
+  )
+
+
+def _parse_accelerator_resource(data: dict[str, Any]) -> _AcceleratorResource:
+  return _AcceleratorResource(
+      acceleratorCount=int(data.get('acceleratorCount', 0)),
+      acceleratorType=str(data.get('acceleratorType', '')),
+  )
+
+
+def _parse_aggregate_reservation(data: dict[str, Any]) -> _AggregateReservation:
+  reserved_resources = [
+      _parse_accelerator_resource(r['accelerator'])
+      for r in data.get('reservedResources', [])
+      if 'accelerator' in r
+  ]
+  in_use_resources = [
+      _parse_accelerator_resource(r['accelerator'])
+      for r in data.get('inUseResources', [])
+      if 'accelerator' in r
+  ]
+  return _AggregateReservation(
+      reservedResources=reserved_resources, inUseResources=in_use_resources
+  )
+
+
+def _parse_reservation(name: str, data: dict[str, Any]) -> _Reservation:
+  specific_reservation = None
+  if 'specificReservation' in data:
+    specific_reservation = _parse_specific_reservation(
+        data['specificReservation']
+    )
+
+  aggregate_reservation = None
+  if 'aggregateReservation' in data:
+    aggregate_reservation = _parse_aggregate_reservation(
+        data['aggregateReservation']
+    )
+
+  deployment_type = data.get('deploymentType', '')
+  resource_policy = data.get('resourcePolicies', {}).get('policy', '')
+
+  return _Reservation(
+      name=name,
+      specificReservation=specific_reservation,
+      aggregateReservation=aggregate_reservation,
+      deployment_type=deployment_type,
+      resource_policy=resource_policy,
+  )
+
+
+def _parse_reservation_sub_block(data: dict[str, Any]) -> _ReservationSubBlock:
+  return _ReservationSubBlock(
+      name=str(data.get('name', '')),
+      count=int(data.get('count', 0)),
+      in_use_count=int(data.get('inUseCount', '0')),
+  )
+
+
+@functools.lru_cache(maxsize=None)
+def _fetch_reservation_from_gcloud(
+    project: str, zone: str, name: str
+) -> _Reservation | None:
+  """Fetches reservation details using gcloud and returns _Reservation object.
+
+  Args:
+    project: Project ID.
+    zone: Zone.
+    name: Reservation name.
+
+  Returns:
+    _Reservation object or None on failure.
+  """
+  command = (
+      f'gcloud beta compute reservations describe {name} '
+      f'--project={project} --zone={zone} '
+      '--format="json(specificReservation,aggregateReservation,status,deploymentType,resourcePolicies)"'
+  )
+  # Basic dry run value to avoid crashes if dry run is enabled globally
+  dry_run_json = json.dumps({
+      'specificReservation': {
+          'count': 100,
+          'inUseCount': 0,
+          'instanceProperties': {},
+      },
+      'status': 'READY',
+      'deploymentType': 'DENSE',
+  })
+
+  return_code, output = run_command_for_value(
+      command,
+      f'Get reservation {name}',
+      dry_run_return_val=dry_run_json,
+  )
+
+  if return_code != 0 or not output.strip():
+    return None
+
+  try:
+    data = json.loads(output)
+    if not data or data.get('status') != 'READY':
+      return None
+    return _parse_reservation(name, data)
+  except (ValueError, IndexError, AttributeError, json.JSONDecodeError) as e:
+    xpk_print(f'Error processing reservation data: {e}. Output: "{output}".')
+    return None
+
+
+def _get_reservation_cached(
+    reservation: ReservationLink,
+) -> _Reservation | None:
+  """Fetches reservation details using gcloud and returns _Reservation object.
+
+  Args:
+    reservation: ReservationLink object.
+
+  Returns:
+    _Reservation object or None on failure.
+  """
+  return _fetch_reservation_from_gcloud(
+      project=reservation.project, zone=reservation.zone, name=reservation.name
+  )
+
+
+_get_reservation_cached.cache_clear = _fetch_reservation_from_gcloud.cache_clear
+
+
 def print_reservations(args) -> int:
   """Print the reservations in the project.
 
@@ -125,70 +316,64 @@ def get_capacity_type(args) -> tuple[CapacityType, int]:
   return capacity_type, return_code
 
 
-def get_reservation_maintenance_interval(reservation: ReservationLink) -> str:
+def get_reservation_maintenance_interval(
+    reservation_link: ReservationLink,
+) -> str:
   """Get reservation maintenance interval.
 
   Args:
-    reservation: reservation object.
+    reservation_link: reservation object.
 
   Returns:
     Maintenance interval as a string.
   """
-  command = (
-      f'gcloud beta compute reservations describe {reservation.name}'
-      f' --project={reservation.project} --zone={reservation.zone} --format="value(specificReservation.instanceProperties.maintenanceInterval)"'
-  )
-  return_code, output = run_command_for_value(
-      command, 'Get reservation maintenance interval'
-  )
-  if return_code != 0:
-    xpk_print(f'Get reservation maintenance interval ERROR {return_code}')
+  reservation = _get_reservation_cached(reservation_link)
+  if not reservation or not reservation.specificReservation:
+    xpk_print(
+        'Get reservation maintenance interval failed for'
+        f' {reservation_link.name}'
+    )
     xpk_exit(1)
-  return output.strip()
 
+  return reservation.specificReservation.maintenance_interval
 
-def get_reservation_placement_policy(reservation: ReservationLink) -> str:
+
+def get_reservation_placement_policy(reservation_link: ReservationLink) -> str:
   """Get reservation placement policy.
 
   Args:
-    reservation: reservation object.
+    reservation_link: reservation object.
 
   Returns:
     Placement policy as a string.
   """
-  command = (
-      f'gcloud beta compute reservations describe {reservation.name}'
-      f' --project={reservation.project} --zone={reservation.zone} --format="value(resourcePolicies.policy)"'
-  )
-  return_code, output = run_command_for_value(
-      command, 'Get reservation placement policy'
-  )
-  if return_code != 0:
-    xpk_print(f'Get reservation placement policy ERROR {return_code}')
+  reservation = _get_reservation_cached(reservation_link)
+  if not reservation:
+    xpk_print(
+        f'Get reservation placement policy failed for {reservation_link.name}'
+    )
     xpk_exit(1)
-  return output.strip()
+
+  return reservation.resource_policy
 
 
-def get_reservation_deployment_type(reservation: ReservationLink) -> str:
+def get_reservation_deployment_type(reservation_link: ReservationLink) -> str:
   """Get reservation deployment type.
 
   Args:
-    reservation: reservation object.
+    reservation_link: reservation object.
 
   Returns:
     Deployment type as a string.
   """
-  command = (
-      f'gcloud beta compute reservations describe {reservation.name}'
-      f' --project={reservation.project} --zone={reservation.zone} --format="value(deploymentType)"'
-  )
-  return_code, output = run_command_for_value(
-      command, 'Get reservation deployment type', dry_run_return_val='DENSE'
-  )
-  if return_code != 0:
-    xpk_print(f'Get reservation deployment type ERROR {return_code}')
+  reservation = _get_reservation_cached(reservation_link)
+  if not reservation:
+    xpk_print(
+        f'Get reservation deployment type failed for {reservation_link.name}'
+    )
     xpk_exit(1)
-  return output.strip()
+
+  return reservation.deployment_type
 
 
 def get_reservations_list(args) -> list[ReservationLink]:
@@ -217,19 +402,13 @@ def verify_reservations_exist(args) -> int:
   Returns:
     0 if successful and 1 otherwise.
   """
-  for reservation in get_reservations_list(args):
-    command = (
-        f'gcloud beta compute reservations describe {reservation.name}'
-        f' --project={reservation.project} --zone={reservation.zone}'
-    )
-    return_code = run_command_with_updates(
-        command, 'Describe reservation', verbose=False
-    )
-    if return_code != 0:
-      xpk_print(f'Describe reservation returned ERROR {return_code}')
+  for reservation_link in get_reservations_list(args):
+    reservation = _get_reservation_cached(reservation_link)
+    if not reservation:
+      xpk_print(f'Describe reservation {reservation_link.name} failed')
       xpk_print(
-          f'Please confirm that your reservation name {reservation.name} is'
-          ' correct.'
+          'Please confirm that your reservation name'
+          f' {reservation_link.name} is correct.'
       )
       return 1
   return 0
@@ -390,3 +569,344 @@ def to_reservation_path(
     if isinstance(reservation, SubBlockReservationLink):
       path += f'/reservationSubBlocks/{reservation.sub_block_name}'
   return path
+
+
+def assess_available_slices(
+    reservations: Sequence[ReservationLink],
+    force_sub_block_targeting: bool,
+    required_hosts: int,
+    system: SystemCharacteristics,
+) -> tuple[list[ReservationCapacity], int]:
+  """Assess the available slices in the reservations.
+
+  Args:
+    reservations: list of reservations to assess.
+    force_sub_block_targeting: if `True`, then the passed `ReservationLink` or `BlockReservationLink` will be flattened to adequate sub-blocks.
+    required_hosts: number of hosts required per slice.
+    system: The system characteristics of the accelerator type.
+
+  Returns:
+    List of capacity reservations with available slices.
+  """
+  reservation_capacities = []
+  for reservation in reservations:
+    capacities, return_code = _assess_available_slices_for_reservation(
+        reservation, force_sub_block_targeting, required_hosts, system
+    )
+    if return_code != 0:
+      return [], 0
+    reservation_capacities.extend(capacities)
+
+  # Deduplicate reservation_capacities, preserving order:
+  reservation_capacities = list(dict.fromkeys(reservation_capacities))
+
+  return reservation_capacities, 0
+
+
+def _get_dry_run_sub_blocks() -> str:
+  """Get dry run sub-blocks based on environment variable."""
+  default_json = '[{"name": "sub0", "count": 16, "inUseCount": 0}]'
+  return os.getenv('DRY_RUN_RESERVATION_SUB_BLOCKS', default_json)
+
+
+def _assess_available_slices_for_reservation(
+    reservation: ReservationLink,
+    force_sub_block_targeting: bool,
+    required_hosts: int,
+    system: SystemCharacteristics,
+) -> tuple[list[ReservationCapacity], int]:
+  """Assess the available slices for a single reservation.
+
+  Args:
+    reservation: reservation to assess.
+    force_sub_block_targeting: if `True`, then the passed `ReservationLink` or `BlockReservationLink` will be flattened to adequate sub-blocks.
+    required_hosts: number of hosts required per slice.
+    system: The system characteristics of the accelerator type.
+
+  Returns:
+    List of available reservations (targeting sub-blocks if applicable).
+  """
+  parent_reservation = _get_reservation_cached(reservation)
+
+  if not parent_reservation:
+    xpk_print(f"WARNING: Failed to fetch reservation '{reservation.name}'.")
+    return [], 0
+
+  if not _verify_reservation_configuration(parent_reservation, system):
+    return [], 0
+
+  if isinstance(reservation, SubBlockReservationLink):
+    available_slices, return_code = _get_available_slices_in_sub_block(
+        reservation, required_hosts
+    )
+    if return_code != 0:
+      return [], 0
+    if available_slices > 0:
+      return [ReservationCapacity(reservation, available_slices)], 0
+    else:
+      xpk_print(
+          f'WARNING: Sub-block {reservation.sub_block_name} is either'
+          ' unhealthy or not fitting. Skipping.'
+      )
+      return [], 0
+
+  if force_sub_block_targeting:
+    if isinstance(reservation, BlockReservationLink):
+      return _get_healthy_and_fitting_sub_blocks_in_block(
+          reservation, required_hosts
+      )
+
+    # reservation instanceof ReservationLink (not Block/SubBlock):
+    blocks, return_code = _get_blocks_in_reservation(reservation)
+    if return_code != 0:
+      return [], 0
+    if blocks:
+      return assess_available_slices(
+          blocks, force_sub_block_targeting, required_hosts, system
+      )
+
+    xpk_print(
+        'WARNING: Super-slicing is enabled, but no blocks found in'
+        f' reservation {reservation.name}. Skipping.'
+    )
+    return [], 0
+
+  count, return_code = _get_reservation_count(
+      parent_reservation, required_hosts, system, reservation
+  )
+  if return_code != 0:
+    return [], return_code
+  return ([ReservationCapacity(reservation, count)] if count > 0 else []), 0
+
+
+def _list_healthy_sub_blocks(
+    reservation: BlockReservationLink | SubBlockReservationLink,
+) -> tuple[list[_ReservationSubBlock], int]:
+  """List healthy sub-blocks for a reservation block or sub-block.
+
+  Args:
+    reservation: The reservation link (must be Block or SubBlock).
+
+  Returns:
+    A tuple containing a list of healthy sub-blocks and the return code.
+  """
+  filter_arg = 'healthInfo.healthStatus=HEALTHY'
+  task_name = f'Count healthy fitting sub-blocks in {reservation.block_name}'
+  dry_run_return_val = _get_dry_run_sub_blocks()
+
+  if isinstance(reservation, SubBlockReservationLink):
+    filter_arg = f'name={reservation.sub_block_name} AND {filter_arg}'
+    task_name = f'Check sub-block {reservation.sub_block_name} health'
+    dry_run_return_val = '[{"name": "sub0", "count": 16, "inUseCount": 0}]'
+
+  command = (
+      f'gcloud beta compute reservations sub-blocks list {reservation.name} '
+      f'--block-name={reservation.block_name} '
+      f'--project={reservation.project} '
+      f'--zone={reservation.zone} '
+      f'--filter="{filter_arg}" '
+      '--format="json(name,count,inUseCount)"'
+  )
+
+  return_code, output = run_command_for_value(
+      command,
+      task_name,
+      dry_run_return_val=dry_run_return_val,
+  )
+
+  if return_code != 0 or not output.strip():
+    return [], return_code
+
+  try:
+    data = json.loads(output)
+    if not data:
+      return [], 0
+    return [_parse_reservation_sub_block(row) for row in data], 0
+  except (ValueError, IndexError, AttributeError, json.JSONDecodeError) as e:
+    xpk_print(f'Error processing sub-block data: {e}. Output: "{output}".')
+    return [], 1
+
+
+def _get_available_slices_in_sub_block(
+    reservation: SubBlockReservationLink,
+    required_hosts: int,
+) -> tuple[int, int]:
+  """Check if a sub-block is healthy and return available slices."""
+  sub_blocks, return_code = _list_healthy_sub_blocks(reservation)
+
+  if return_code != 0 or not sub_blocks:
+    return 0, return_code
+
+  assert len(sub_blocks) == 1
+  sub_block = sub_blocks[0]
+  available_slices = (
+      sub_block.count - sub_block.in_use_count
+  ) // required_hosts
+  return available_slices, 0
+
+
+def _get_healthy_and_fitting_sub_blocks_in_block(
+    reservation: BlockReservationLink,
+    required_hosts: int,
+) -> tuple[list[ReservationCapacity], int]:
+  """Get healthy and fitting sub-blocks in a block."""
+  sub_blocks, return_code = _list_healthy_sub_blocks(reservation)
+
+  if return_code != 0:
+    return [], return_code
+
+  available_capacities: list[ReservationCapacity] = [
+      ReservationCapacity(
+          SubBlockReservationLink(
+              project=reservation.project,
+              name=reservation.name,
+              zone=reservation.zone,
+              block_name=reservation.block_name,
+              sub_block_name=sub_block.name,
+          ),
+          available_slices=(sub_block.count - sub_block.in_use_count)
+          // required_hosts,
+      )
+      for sub_block in sub_blocks
+  ]
+  return available_capacities, 0
+
+
+def _get_blocks_in_reservation(
+    reservation: ReservationLink,
+) -> tuple[list[BlockReservationLink], int]:
+  """Get blocks in a reservation."""
+  command = (
+      f'gcloud beta compute reservations blocks list {reservation.name} '
+      f'--project={reservation.project} '
+      f'--zone={reservation.zone} '
+      '--format="value(name)"'
+  )
+  return_code, output = run_command_for_value(
+      command,
+      f'Get blocks in reservation {reservation.name}',
+      dry_run_return_val='block0',
+  )
+  if return_code != 0:
+    xpk_print(
+        f'Get blocks in reservation {reservation.name} failed with'
+        f' {return_code}'
+    )
+    return [], return_code
+
+  return [
+      BlockReservationLink(
+          project=reservation.project,
+          name=reservation.name,
+          zone=reservation.zone,
+          block_name=name,
+      )
+      for name in output.strip().splitlines()
+      if name
+  ], 0
+
+
+def _calculate_target_accelerator_type(
+    link: ReservationLink, system: SystemCharacteristics
+) -> str:
+  reservation_accelerator_type = system.reservation_accelerator_type
+  assert reservation_accelerator_type
+
+  if system.accelerator_type == AcceleratorType.TPU:
+    project_number = project_id_to_project_number(link.project)
+    return (
+        f'projects/{project_number}/zones/{link.zone}/acceleratorTypes/'
+        f'{reservation_accelerator_type}'
+    )
+  else:
+    return reservation_accelerator_type
+
+
+def _verify_reservation_configuration(
+    reservation: _Reservation, system: SystemCharacteristics
+) -> bool:
+  """Checks if the reservation matches the system requirements.
+
+  Args:
+    reservation: The reservation object.
+    system: The system characteristics.
+
+  Returns:
+    True if valid, False otherwise. Prints error message on failure.
+  """
+  if not reservation.specificReservation:
+    return True
+
+  if is_dry_run() and not reservation.specificReservation.machine_type:
+    return True
+
+  if system.accelerator_type == AcceleratorType.TPU:
+    if reservation.specificReservation.machine_type != system.gce_machine_type:
+      xpk_print(
+          f"ERROR: Reservation '{reservation.name}' has machine type"
+          f" '{reservation.specificReservation.machine_type}', but requested"
+          f" system requires '{system.gce_machine_type}'."
+      )
+      return False
+  elif system.accelerator_type == AcceleratorType.GPU:
+    target_accel = system.reservation_accelerator_type
+    has_matching_accelerator = any(
+        acc.acceleratorType == target_accel
+        for acc in reservation.specificReservation.guest_accelerators
+    )
+    if not has_matching_accelerator:
+      xpk_print(
+          f"ERROR: Reservation '{reservation.name}' does not have a matching"
+          f" guest accelerator for '{target_accel}'."
+      )
+      return False
+  return True
+
+
+def _get_reservation_count(
+    reservation: _Reservation,
+    required_hosts: int,
+    system: SystemCharacteristics,
+    link: ReservationLink,
+) -> tuple[int, int]:
+  """Get capacity count of a reservation.
+
+  Args:
+    reservation: The reservation object.
+    required_hosts: number of hosts required per slice.
+    system: The system characteristics of the accelerator type.
+    link: The reservation link (for logging/identification).
+
+  Returns:
+    Number of available slots in the reservation.
+  """
+  count = 0
+  in_use_count = 0
+
+  if reservation.specificReservation:
+    count = int(reservation.specificReservation.count)
+    in_use_count = int(reservation.specificReservation.inUseCount)
+  elif reservation.aggregateReservation:
+    reserved_resources = reservation.aggregateReservation.reservedResources
+    target_accelerator_type = _calculate_target_accelerator_type(link, system)
+    count = next(
+        (
+            r.acceleratorCount
+            for r in reserved_resources
+            if r.acceleratorType == target_accelerator_type
+        ),
+        0,
+    )
+
+    in_use_resources = reservation.aggregateReservation.inUseResources
+    in_use_count = next(
+        (
+            r.acceleratorCount
+            for r in in_use_resources
+            if r.acceleratorType == target_accelerator_type
+        ),
+        0,
+    )
+
+  available_hosts = max(0, count - in_use_count)
+  return available_hosts // required_hosts, 0
diff --git a/src/xpk/core/capacity_test.py b/src/xpk/core/capacity_test.py
index e217e4b19..d6b47cdac 100644
--- a/src/xpk/core/capacity_test.py
+++ b/src/xpk/core/capacity_test.py
@@ -15,54 +15,144 @@
 """
 
 import pytest
+import sys
 from unittest.mock import MagicMock, patch
+
+# Mock google.cloud.resourcemanager_v3 before importing capacity
+sys.modules['google.cloud.resourcemanager_v3'] = MagicMock()
+
 from .capacity import (
     get_reservation_deployment_type,
+    get_reservation_placement_policy,
+    get_reservation_maintenance_interval,
     parse_reservation,
-    ReservationLink,
-    BlockReservationLink,
-    SubBlockReservationLink,
     get_capacity_type,
     CapacityType,
     verify_reservations_exist,
     get_reservations_list,
+    assess_available_slices,
     to_reservation_path,
+    ReservationLink,
+    BlockReservationLink,
+    SubBlockReservationLink,
+    ReservationCapacity,
+    _parse_reservation,
+    _parse_reservation_sub_block,
+    _SpecificReservation,
+    _AcceleratorResource,
+    _verify_reservation_configuration,
+    _Reservation,
+    _get_reservation_cached,
 )
+from xpk.core.testing.commands_tester import CommandsTester
+from .system_characteristics import SystemCharacteristics, AcceleratorType, DockerPlatform, GpuConfig
 
 
-@patch('xpk.core.capacity.xpk_print')
-def test_get_reservation_deployment_type_exits_with_command_fails(
-    xpk_print: MagicMock, mocker
-):
-  mocker.patch(
-      target='xpk.core.capacity.run_command_for_value', return_value=(1, '')
+@pytest.fixture
+def commands_tester(mocker):
+  return CommandsTester(mocker)
+
+
+@pytest.fixture(autouse=True)
+def clear_capacity_cache():
+  _get_reservation_cached.cache_clear()
+  yield
+  _get_reservation_cached.cache_clear()
+
+
+@pytest.fixture
+def test_system():
+  return SystemCharacteristics(
+      topology='2x2x1',
+      vms_per_slice=1,
+      gke_accelerator='test-accel',
+      gce_machine_type='test-machine',
+      chips_per_vm=1,
+      accelerator_type=AcceleratorType.TPU,
+      device_type='test-device',
+      supports_sub_slicing=False,
+      supports_super_slicing=False,
+      supports_accelerator_network_profile=False,
+      docker_platform=DockerPlatform.AMD,
   )
-  with pytest.raises(SystemExit):
-    get_reservation_deployment_type(
-        reservation=ReservationLink(
-            project='project', name='reservation', zone='zone'
-        )
-    )
 
-  assert (
-      'Get reservation deployment type ERROR 1'
-      in xpk_print.mock_calls[0].args[0]
+
+@patch('xpk.core.capacity._get_reservation_cached')
+def test_get_reservation_deployment_type_uses_cached(mock_get_cached):
+  mock_res = MagicMock(spec=_Reservation)
+  mock_res.deployment_type = 'DENSE'
+  mock_get_cached.return_value = mock_res
+
+  res_link = ReservationLink(project='p', name='r', zone='z')
+
+  result = get_reservation_deployment_type(res_link)
+
+  assert result == 'DENSE'
+  mock_get_cached.assert_called_once_with(
+      ReservationLink(project='p', zone='z', name='r')
   )
 
 
-def test_get_reservation_deployment_type_returns_deployment_type_when_command_succeeds(
-    mocker,
-):
-  mocker.patch(
-      target='xpk.core.capacity.run_command_for_value',
-      return_value=(0, 'DENSE'),
+@patch('xpk.core.capacity._get_reservation_cached')
+def test_get_reservation_placement_policy_uses_cached(mock_get_cached):
+  mock_res = MagicMock(spec=_Reservation)
+  mock_res.resource_policy = 'compact'
+  mock_get_cached.return_value = mock_res
+
+  res_link = ReservationLink(project='p', name='r', zone='z')
+
+  result = get_reservation_placement_policy(res_link)
+
+  assert result == 'compact'
+  mock_get_cached.assert_called_once_with(
+      ReservationLink(project='p', zone='z', name='r')
   )
-  result = get_reservation_deployment_type(
-      reservation=ReservationLink(
-          project='project', name='reservation', zone='zone'
-      )
+
+
+@patch('xpk.core.capacity._get_reservation_cached')
+def test_get_reservation_maintenance_interval_uses_cached(mock_get_cached):
+  mock_res = MagicMock(spec=_Reservation)
+  mock_res.specificReservation = MagicMock(spec=_SpecificReservation)
+  mock_res.specificReservation.maintenance_interval = 'PERIODIC'
+  mock_get_cached.return_value = mock_res
+
+  res_link = ReservationLink(project='p', name='r', zone='z')
+
+  result = get_reservation_maintenance_interval(res_link)
+
+  assert result == 'PERIODIC'
+  mock_get_cached.assert_called_once_with(
+      ReservationLink(project='p', zone='z', name='r')
   )
-  assert result == 'DENSE'
+
+
+@patch('xpk.core.capacity._get_reservation_cached')
+def test_verify_reservations_exist_uses_cached_success(mock_get_cached, mocker):
+  mock_get_cached.return_value = MagicMock(spec=_Reservation)
+
+  args = mocker.Mock(reservation='r1,r2', project='p', zone='z')
+
+  assert verify_reservations_exist(args) == 0
+  assert mock_get_cached.call_count == 2
+  mock_get_cached.assert_any_call(
+      ReservationLink(project='p', zone='z', name='r1')
+  )
+  mock_get_cached.assert_any_call(
+      ReservationLink(project='p', zone='z', name='r2')
+  )
+
+
+@patch('xpk.core.capacity._get_reservation_cached')
+def test_verify_reservations_exist_uses_cached_failure(mock_get_cached, mocker):
+  mock_get_cached.side_effect = [
+      MagicMock(),
+      None,
+  ]  # First exists, second fails
+
+  args = mocker.Mock(reservation='r1,r2', project='p', zone='z')
+
+  assert verify_reservations_exist(args) == 1
+  assert mock_get_cached.call_count == 2
 
 
 @pytest.mark.parametrize(
@@ -209,18 +299,13 @@ def test_get_capacity_type_multiple_reservations(mocker):
       project='project',
       zone='zone',
   )
+
   capacity_type, return_code = get_capacity_type(args)
+
   assert capacity_type == CapacityType.RESERVATION
   assert return_code == 0
 
 
-@patch('xpk.core.capacity.run_command_with_updates', return_value=0)
-def test_verify_reservations_exist_multiple(mock_run, mocker):
-  args = mocker.Mock(reservation='res1,res2', project='project', zone='zone')
-  assert verify_reservations_exist(args) == 0
-  assert mock_run.call_count == 2
-
-
 def test_get_reservations_list_with_single_reservation(mocker):
   args = mocker.Mock(reservation='res1', project='project', zone='zone')
   assert get_reservations_list(args) == [
@@ -252,3 +337,1136 @@ def test_get_reservations_list_none(mocker):
 def test_get_reservations_list_empty(mocker):
   args = mocker.Mock(reservation='')
   assert get_reservations_list(args) == []
+
+
+def test_to_reservation_path_sub_block():
+  res = SubBlockReservationLink(
+      project='project',
+      name='reservation',
+      zone='zone',
+      block_name='block',
+      sub_block_name='sub-block',
+  )
+  assert (
+      to_reservation_path(res, 'project')
+      == 'reservation/reservationBlocks/block/reservationSubBlocks/sub-block'
+  )
+
+
+def test_to_reservation_path_block():
+  res_block = BlockReservationLink(
+      project='project',
+      name='reservation',
+      zone='zone',
+      block_name='block',
+  )
+  assert (
+      to_reservation_path(res_block, 'project')
+      == 'reservation/reservationBlocks/block'
+  )
+
+
+def test_to_reservation_path_reservation():
+  res_simple = ReservationLink(
+      project='project', name='reservation', zone='zone'
+  )
+  assert to_reservation_path(res_simple, 'project') == 'reservation'
+
+
+def test_assess_available_slices_sub_block_healthy(
+    commands_tester: CommandsTester,
+    test_system: SystemCharacteristics,
+):
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 48, "inUseCount": 2,'
+              ' "instanceProperties": {"machineType": "test-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+  commands_tester.set_result_for_command(
+      (0, '[{"count": 1, "inUseCount": 0}]'),
+      'gcloud beta compute reservations sub-blocks list',
+  )
+  res = SubBlockReservationLink(
+      project='project',
+      name='reservation',
+      zone='zone',
+      block_name='block',
+      sub_block_name='sub-block',
+  )
+
+  slices, return_code = assess_available_slices(
+      [res],
+      force_sub_block_targeting=False,
+      required_hosts=1,
+      system=test_system,
+  )
+
+  assert slices == [ReservationCapacity(res, 1)]
+  assert return_code == 0
+
+
+def test_assess_available_slices_sub_block_unhealthy(
+    commands_tester: CommandsTester,
+    test_system: SystemCharacteristics,
+):
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 48, "inUseCount": 2,'
+              ' "instanceProperties": {"machineType": "test-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+  commands_tester.set_result_for_command(
+      (0, '[]'), 'gcloud beta compute reservations sub-blocks list'
+  )
+  res = SubBlockReservationLink(
+      project='project',
+      name='reservation',
+      zone='zone',
+      block_name='block',
+      sub_block_name='sub-block',
+  )
+  slices, return_code = assess_available_slices(
+      [res],
+      force_sub_block_targeting=False,
+      required_hosts=1,
+      system=test_system,
+  )
+
+  assert not slices
+  assert return_code == 0
+
+
+def test_assess_available_slices_block_healthy(
+    commands_tester: CommandsTester, test_system: SystemCharacteristics
+):
+  # Mock describe
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 48, "inUseCount": 2,'
+              ' "instanceProperties": {"machineType": "test-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+  # Mock 2 healthy sub-blocks
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '[{"name": "sub1", "count": 1, "inUseCount": 0}, {"name":'
+              ' "sub2", "count": 1, "inUseCount": 0}]'
+          ),
+      ),
+      'gcloud beta compute reservations sub-blocks list',
+  )
+  res = BlockReservationLink(
+      project='project',
+      name='reservation',
+      zone='zone',
+      block_name='block',
+  )
+
+  slices, return_code = assess_available_slices(
+      [res],
+      force_sub_block_targeting=True,
+      required_hosts=1,
+      system=test_system,
+  )
+
+  assert return_code == 0
+  assert slices == [
+      ReservationCapacity(
+          SubBlockReservationLink(
+              project='project',
+              name='reservation',
+              zone='zone',
+              block_name='block',
+              sub_block_name='sub1',
+          ),
+          available_slices=1,
+      ),
+      ReservationCapacity(
+          SubBlockReservationLink(
+              project='project',
+              name='reservation',
+              zone='zone',
+              block_name='block',
+              sub_block_name='sub2',
+          ),
+          available_slices=1,
+      ),
+  ]
+
+
+def test_assess_available_slices_block_unhealthy(
+    commands_tester: CommandsTester,
+    test_system: SystemCharacteristics,
+):
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 48, "inUseCount": 2,'
+              ' "instanceProperties": {"machineType": "test-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+  commands_tester.set_result_for_command(
+      (0, '[]'), 'gcloud beta compute reservations sub-blocks list'
+  )
+  res = BlockReservationLink(
+      project='project',
+      name='reservation',
+      zone='zone',
+      block_name='block',
+  )
+
+  slices, return_code = assess_available_slices(
+      [res],
+      force_sub_block_targeting=True,
+      required_hosts=1,
+      system=test_system,
+  )
+
+  assert not slices
+  assert return_code == 0
+
+
+def test_assess_available_slices_link_with_blocks(
+    commands_tester: CommandsTester,
+    test_system: SystemCharacteristics,
+):
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 48, "inUseCount": 2,'
+              ' "instanceProperties": {"machineType": "test-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+  commands_tester.set_result_for_command(
+      (0, 'block1'), 'gcloud beta compute reservations blocks list'
+  )
+  commands_tester.set_result_for_command(
+      (0, '[{"name": "sub1", "count": 1, "inUseCount": 0}]'),
+      'gcloud beta compute reservations sub-blocks list',
+      '--block-name=block1',
+  )
+
+  res = ReservationLink(project='project', name='reservation', zone='zone')
+  slices, return_code = assess_available_slices(
+      [res],
+      force_sub_block_targeting=True,
+      required_hosts=1,
+      system=test_system,
+  )
+
+  assert return_code == 0
+  assert slices == [
+      ReservationCapacity(
+          SubBlockReservationLink(
+              project='project',
+              name='reservation',
+              zone='zone',
+              block_name='block1',
+              sub_block_name='sub1',
+          ),
+          available_slices=1,
+      )
+  ]
+
+
+def test_assess_available_slices_link_without_blocks(
+    commands_tester: CommandsTester,
+    test_system: SystemCharacteristics,
+):
+  commands_tester.set_result_for_command(
+      (0, ''), 'gcloud beta compute reservations blocks list'
+  )
+  # Mock getting count
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 2, "inUseCount": 0,'
+              ' "instanceProperties": {"machineType": "test-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+
+  res = ReservationLink(project='project', name='reservation', zone='zone')
+  slices, return_code = assess_available_slices(
+      [res],
+      force_sub_block_targeting=False,
+      required_hosts=1,
+      system=test_system,
+  )
+  assert return_code == 0
+  assert slices == [
+      ReservationCapacity(
+          ReservationLink(project='project', name='reservation', zone='zone'),
+          available_slices=2,
+      )
+  ]
+
+
+def test_assess_available_slices_link_without_blocks_sub_block_targeting(
+    commands_tester: CommandsTester,
+    test_system: SystemCharacteristics,
+):
+  commands_tester.set_result_for_command(
+      (0, ''), 'gcloud beta compute reservations blocks list'
+  )
+  # Mock getting count
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 2, "inUseCount": 0,'
+              ' "instanceProperties": {"machineType": "test-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+
+  res = ReservationLink(project='project', name='reservation', zone='zone')
+  slices, return_code = assess_available_slices(
+      [res],
+      force_sub_block_targeting=True,
+      required_hosts=1,
+      system=test_system,
+  )
+  assert return_code == 0
+  assert not slices
+
+
+def test_assess_available_slices_host_filtering_insufficient_hosts(
+    commands_tester: CommandsTester,
+    test_system: SystemCharacteristics,
+):
+  # Mock a sub-block that has 14 free hosts but we need 16
+  commands_tester.set_result_for_command(
+      (0, '[{"count": 16, "inUseCount": 2}]'),
+      'gcloud beta compute reservations sub-blocks list',
+  )
+  res = SubBlockReservationLink(
+      project='project',
+      name='reservation',
+      zone='zone',
+      block_name='block',
+      sub_block_name='sub-block',
+  )
+
+  slices, return_code = assess_available_slices(
+      [res],
+      force_sub_block_targeting=False,
+      required_hosts=16,
+      system=test_system,
+  )
+
+  assert not slices
+  assert return_code == 0
+
+
+def test_assess_available_slices_host_filtering_sufficient_hosts(
+    commands_tester: CommandsTester,
+    test_system: SystemCharacteristics,
+):
+  # Mock a reservation that has 46 free hosts, and we need 16 per slice.
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 48, "inUseCount": 2,'
+              ' "instanceProperties": {"machineType": "test-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+  res_link = ReservationLink(project='p', name='r', zone='z')
+
+  slices, return_code = assess_available_slices(
+      [res_link],
+      force_sub_block_targeting=False,
+      required_hosts=16,
+      system=test_system,
+  )
+
+  assert return_code == 0
+  assert slices == [
+      ReservationCapacity(
+          ReservationLink(project='p', name='r', zone='z'), available_slices=2
+      )
+  ]
+
+
+@patch('xpk.core.capacity.project_id_to_project_number', return_value='12345')
+def test_assess_available_slices_aggregate_reservation(
+    mock_project_id,
+    commands_tester: CommandsTester,
+    test_system: SystemCharacteristics,
+):
+  # For TPU, target type includes project number and zone
+  target_type = f'projects/12345/zones/zone/acceleratorTypes/{test_system.reservation_accelerator_type}'
+  json_output = f"""
+  {{
+      "aggregateReservation": {{
+          "reservedResources": [
+              {{
+                  "accelerator": {{
+                      "acceleratorType": "{target_type}",
+                      "acceleratorCount": 100
+                  }}
+              }},
+              {{
+                  "accelerator": {{
+                      "acceleratorType": "wrong-type",
+                      "acceleratorCount": 100
+                  }}
+              }}
+          ],
+          "inUseResources": [
+              {{
+                  "accelerator": {{
+                      "acceleratorType": "{target_type}",
+                      "acceleratorCount": 20
+                  }}
+              }},
+              {{
+                  "accelerator": {{
+                      "acceleratorType": "accelerator-2",
+                      "acceleratorCount": 50
+                  }}
+              }}
+          ]
+      }},
+      "status": "READY"
+  }}
+  """
+  commands_tester.set_result_for_command(
+      (0, json_output),
+      'gcloud beta compute reservations describe',
+  )
+  res = ReservationLink(project='project', name='reservation', zone='zone')
+
+  slices, return_code = assess_available_slices(
+      [res],
+      force_sub_block_targeting=False,
+      required_hosts=1,
+      system=test_system,
+  )
+
+  assert return_code == 0
+  assert slices == [
+      ReservationCapacity(
+          ReservationLink(project='project', name='reservation', zone='zone'),
+          available_slices=80,
+      )
+  ]
+
+
+def test_assess_available_slices_failures_sub_block_check(
+    commands_tester: CommandsTester,
+    test_system: SystemCharacteristics,
+):
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 100, "inUseCount": 0,'
+              ' "instanceProperties": {"machineType": "test-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+  res_sub = SubBlockReservationLink(
+      project='project',
+      name='reservation',
+      zone='zone',
+      block_name='block',
+      sub_block_name='sub-block',
+  )
+  commands_tester.set_result_for_command(
+      (1, ''), 'gcloud beta compute reservations sub-blocks list'
+  )
+
+  slices, return_code = assess_available_slices(
+      [res_sub],
+      force_sub_block_targeting=False,
+      required_hosts=1,
+      system=test_system,
+  )
+
+  assert not slices
+  assert return_code == 0
+
+
+def test_assess_available_slices_failures_block_sub_blocks_check(
+    commands_tester: CommandsTester,
+    test_system: SystemCharacteristics,
+):
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 100, "inUseCount": 0,'
+              ' "instanceProperties": {"machineType": "test-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+  res_block = BlockReservationLink(
+      project='project',
+      name='reservation',
+      zone='zone',
+      block_name='block',
+  )
+  commands_tester.set_result_for_command(
+      (1, ''), 'gcloud beta compute reservations sub-blocks list'
+  )
+
+  slices, return_code = assess_available_slices(
+      [res_block],
+      force_sub_block_targeting=True,
+      required_hosts=1,
+      system=test_system,
+  )
+
+  assert not slices
+  assert return_code == 0
+
+
+def test_assess_available_slices_failures_reservation_blocks_check(
+    commands_tester: CommandsTester,
+    test_system: SystemCharacteristics,
+):
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 100, "inUseCount": 0,'
+              ' "instanceProperties": {"machineType": "test-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+  res = ReservationLink(project='project', name='reservation', zone='zone')
+  commands_tester.set_result_for_command(
+      (1, ''), 'gcloud beta compute reservations blocks list'
+  )
+
+  slices, return_code = assess_available_slices(
+      [res],
+      force_sub_block_targeting=True,
+      required_hosts=1,
+      system=test_system,
+  )
+
+  assert not slices
+  assert return_code == 0
+
+
+def test_assess_available_slices_failures_reservation_count_check(
+    commands_tester: CommandsTester,
+    test_system: SystemCharacteristics,
+):
+  res = ReservationLink(project='project', name='reservation', zone='zone')
+  commands_tester.set_result_for_command(
+      (1, ''), 'gcloud beta compute reservations describe'
+  )
+
+  slices, return_code = assess_available_slices(
+      [res],
+      force_sub_block_targeting=False,
+      required_hosts=1,
+      system=test_system,
+  )
+
+  assert not slices
+  assert return_code == 0
+
+
+def test_assess_available_slices_mixed_reservations_with_subblock_targeting(
+    commands_tester: CommandsTester,
+    test_system: SystemCharacteristics,
+):
+  # Mock describe for all reservations
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 48, "inUseCount": 2,'
+              ' "instanceProperties": {"machineType": "test-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+
+  # Mock block reservation with 2 healthy sub-blocks
+
+  block_res = BlockReservationLink(
+      project='project', name='res1', zone='zone', block_name='block1'
+  )
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '[{"name": "sub1", "count": 1, "inUseCount": 0}, {"name":'
+              ' "sub2", "count": 1, "inUseCount": 0}]'
+          ),
+      ),
+      'gcloud beta compute reservations sub-blocks list res1',
+      '--block-name=block1',
+  )
+
+  # Mock healthy sub-block reservation
+  sub_res_healthy = SubBlockReservationLink(
+      project='project',
+      name='res2',
+      zone='zone',
+      block_name='block2',
+      sub_block_name='sub3',
+  )
+  commands_tester.set_result_for_command(
+      (0, '[{"count": 1, "inUseCount": 0}]'),
+      'gcloud beta compute reservations sub-blocks list res2',
+      '--filter="name=sub3 AND healthInfo.healthStatus=HEALTHY"',
+  )
+
+  # Mock unhealthy sub-block reservation
+  sub_res_unhealthy = SubBlockReservationLink(
+      project='project',
+      name='res3',
+      zone='zone',
+      block_name='block3',
+      sub_block_name='sub4',
+  )
+  commands_tester.set_result_for_command(
+      (0, '[]'),
+      'gcloud beta compute reservations sub-blocks list res3',
+      '--filter="name=sub4 AND healthInfo.healthStatus=HEALTHY"',
+  )
+
+  slices, return_code = assess_available_slices(
+      [block_res, sub_res_healthy, sub_res_unhealthy],
+      force_sub_block_targeting=True,
+      required_hosts=1,
+      system=test_system,
+  )
+
+  assert return_code == 0
+  assert slices == [
+      ReservationCapacity(
+          SubBlockReservationLink(
+              project='project',
+              name='res1',
+              zone='zone',
+              block_name='block1',
+              sub_block_name='sub1',
+          ),
+          available_slices=1,
+      ),
+      ReservationCapacity(
+          SubBlockReservationLink(
+              project='project',
+              name='res1',
+              zone='zone',
+              block_name='block1',
+              sub_block_name='sub2',
+          ),
+          available_slices=1,
+      ),
+      ReservationCapacity(
+          sub_res_healthy,
+          available_slices=1,
+      ),
+  ]
+
+
+def test_assess_available_slices_deduplicates(
+    commands_tester: CommandsTester, test_system: SystemCharacteristics
+):
+  # Mock describe
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 48, "inUseCount": 2,'
+              ' "instanceProperties": {"machineType": "test-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+
+  block_res = BlockReservationLink(
+      project='project', name='res1', zone='zone', block_name='block1'
+  )
+  sub_block_name = 'sub1'
+  commands_tester.set_result_for_command(
+      (0, f'[{{"name": "{sub_block_name}", "count": 1, "inUseCount": 0}}]'),
+      'gcloud beta compute reservations sub-blocks list res1',
+      '--block-name=block1',
+  )
+  sub_res = SubBlockReservationLink(
+      project='project',
+      name='res1',
+      zone='zone',
+      block_name='block1',
+      sub_block_name=sub_block_name,
+  )
+  commands_tester.set_result_for_command(
+      (0, '[{"count": 1, "inUseCount": 0}]'),
+      'gcloud beta compute reservations sub-blocks list res1',
+      '--block-name=block1',
+      f'--filter="name={sub_block_name}',
+  )
+
+  slices, return_code = assess_available_slices(
+      [block_res, sub_res],
+      force_sub_block_targeting=True,
+      required_hosts=1,
+      system=test_system,
+  )
+
+  assert return_code == 0
+  assert slices == [
+      ReservationCapacity(
+          SubBlockReservationLink(
+              project='project',
+              name='res1',
+              zone='zone',
+              block_name='block1',
+              sub_block_name='sub1',
+          ),
+          available_slices=1,
+      )
+  ]
+
+
+def test_parse_specific_reservation():
+  data = {
+      'specificReservation': {
+          'count': '10',
+          'inUseCount': '2',
+          'instanceProperties': {
+              'machineType': 'test-machine',
+              'guestAccelerators': [{
+                  'acceleratorCount': 1,
+                  'acceleratorType': 'nvidia-test',
+              }],
+          },
+      },
+      'status': 'READY',
+  }
+  res = _parse_reservation('res1', data)
+  assert res.name == 'res1'
+  assert res.specificReservation == _SpecificReservation(
+      count=10,
+      inUseCount=2,
+      machine_type='test-machine',
+      guest_accelerators=[
+          _AcceleratorResource(
+              acceleratorCount=1, acceleratorType='nvidia-test'
+          )
+      ],
+  )
+  assert res.aggregateReservation is None
+
+
+def test_parse_specific_reservation_defaults():
+  data = {'specificReservation': {}, 'status': 'READY'}
+  res = _parse_reservation('res1', data)
+  assert res.specificReservation == _SpecificReservation(
+      count=0, inUseCount=0, machine_type='', guest_accelerators=[]
+  )
+
+
+def test_parse_aggregate_reservation():
+  data = {
+      'aggregateReservation': {
+          'reservedResources': [{
+              'accelerator': {
+                  'acceleratorCount': 100,
+                  'acceleratorType': 'tpu',
+              }
+          }],
+          'inUseResources': [{
+              'accelerator': {
+                  'acceleratorCount': 20,
+                  'acceleratorType': 'tpu',
+              }
+          }],
+      },
+      'status': 'READY',
+  }
+  res = _parse_reservation('res1', data)
+  assert res.aggregateReservation is not None
+  assert len(res.aggregateReservation.reservedResources) == 1
+  assert res.aggregateReservation.reservedResources[0] == _AcceleratorResource(
+      acceleratorCount=100, acceleratorType='tpu'
+  )
+  assert len(res.aggregateReservation.inUseResources) == 1
+  assert res.aggregateReservation.inUseResources[0] == _AcceleratorResource(
+      acceleratorCount=20, acceleratorType='tpu'
+  )
+
+
+def test_parse_reservation_sub_block():
+  data = {'name': 'sub1', 'count': 10, 'inUseCount': 2}
+  res = _parse_reservation_sub_block(data)
+  assert res.name == 'sub1'
+  assert res.count == 10
+  assert res.in_use_count == 2
+
+
+def test_parse_reservation_sub_block_defaults():
+  data = {}
+  res = _parse_reservation_sub_block(data)
+  assert res.name == ''
+  assert res.count == 0
+  assert res.in_use_count == 0
+
+
+def test_get_reservation_count_validates_tpu_machine_type(
+    commands_tester: CommandsTester, test_system: SystemCharacteristics
+):
+  # Success case: matches
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 10, "inUseCount": 2,'
+              ' "instanceProperties": {"machineType": "test-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+  res_link = ReservationLink(project='p', name='r', zone='z')
+  count, return_code = assess_available_slices(
+      [res_link],
+      force_sub_block_targeting=False,
+      required_hosts=1,
+      system=test_system,
+  )
+  assert return_code == 0
+  assert count[0].available_slices == 8
+
+  # Failure case: mismatch
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 10, "inUseCount": 2,'
+              ' "instanceProperties": {"machineType": "wrong-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+  res_link_fail = ReservationLink(project='p', name='r-fail', zone='z')
+  count, return_code = assess_available_slices(
+      [res_link_fail],
+      force_sub_block_targeting=False,
+      required_hosts=1,
+      system=test_system,
+  )
+  assert return_code == 0
+  assert not count
+
+
+def test_get_reservation_count_validates_gpu_accelerator_type(
+    commands_tester: CommandsTester,
+):
+  gpu_system = SystemCharacteristics(
+      topology='N/A',
+      vms_per_slice=1,
+      gke_accelerator='nvidia-test',
+      gce_machine_type='g2-standard-12',
+      chips_per_vm=1,
+      accelerator_type=AcceleratorType.GPU,
+      device_type='test-gpu',
+      supports_sub_slicing=False,
+      supports_super_slicing=False,
+      supports_accelerator_network_profile=False,
+      docker_platform=DockerPlatform.AMD,
+      gpu_config=GpuConfig(requires_topology=False),
+  )
+
+  # Success case: matches
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 10, "inUseCount": 2,'
+              ' "instanceProperties": {"guestAccelerators":'
+              ' [{"acceleratorType": "nvidia-test"}]}}, "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+  res_link = ReservationLink(project='p', name='r', zone='z')
+  count, return_code = assess_available_slices(
+      [res_link],
+      force_sub_block_targeting=False,
+      required_hosts=1,
+      system=gpu_system,
+  )
+  assert return_code == 0
+  assert count[0].available_slices == 8
+
+  # Failure case: mismatch
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 10, "inUseCount": 2,'
+              ' "instanceProperties": {"guestAccelerators":'
+              ' [{"acceleratorType": "nvidia-wrong"}]}}, "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+  res_link_fail = ReservationLink(project='p', name='r-fail', zone='z')
+  count, return_code = assess_available_slices(
+      [res_link_fail],
+      force_sub_block_targeting=False,
+      required_hosts=1,
+      system=gpu_system,
+  )
+  assert return_code == 0
+  assert not count
+
+
+@patch('xpk.core.capacity.xpk_print')
+def test_verify_reservation_configuration(xpk_print: MagicMock):
+  # 1. Valid TPU
+  valid_tpu_system = SystemCharacteristics(
+      topology='2x2x1',
+      vms_per_slice=1,
+      gke_accelerator='test-accel',
+      gce_machine_type='test-machine',
+      chips_per_vm=1,
+      accelerator_type=AcceleratorType.TPU,
+      device_type='test-device',
+      supports_sub_slicing=False,
+      supports_super_slicing=False,
+      supports_accelerator_network_profile=False,
+      docker_platform=DockerPlatform.AMD,
+  )
+  valid_tpu_reservation = _Reservation(
+      name='res1',
+      specificReservation=_SpecificReservation(
+          count=10,
+          inUseCount=0,
+          machine_type='test-machine',
+          guest_accelerators=[],
+      ),
+      aggregateReservation=None,
+  )
+  assert _verify_reservation_configuration(
+      valid_tpu_reservation, valid_tpu_system
+  )
+
+  # 2. Invalid TPU (machine type mismatch)
+  invalid_tpu_reservation = _Reservation(
+      name='res1',
+      specificReservation=_SpecificReservation(
+          count=10,
+          inUseCount=0,
+          machine_type='wrong-machine',
+          guest_accelerators=[],
+      ),
+      aggregateReservation=None,
+  )
+  assert not _verify_reservation_configuration(
+      invalid_tpu_reservation, valid_tpu_system
+  )
+  assert 'ERROR: Reservation' in xpk_print.call_args[0][0]
+
+  # 3. Valid GPU
+  valid_gpu_system = SystemCharacteristics(
+      topology='N/A',
+      vms_per_slice=1,
+      gke_accelerator='nvidia-test',
+      gce_machine_type='g2-standard-12',
+      chips_per_vm=1,
+      accelerator_type=AcceleratorType.GPU,
+      device_type='test-gpu',
+      supports_sub_slicing=False,
+      supports_super_slicing=False,
+      supports_accelerator_network_profile=False,
+      docker_platform=DockerPlatform.AMD,
+      gpu_config=GpuConfig(requires_topology=False),
+  )
+  valid_gpu_reservation = _Reservation(
+      name='res2',
+      specificReservation=_SpecificReservation(
+          count=10,
+          inUseCount=0,
+          machine_type='g2-standard-12',
+          guest_accelerators=[
+              _AcceleratorResource(
+                  acceleratorCount=1, acceleratorType='nvidia-test'
+              )
+          ],
+      ),
+      aggregateReservation=None,
+  )
+  assert _verify_reservation_configuration(
+      valid_gpu_reservation, valid_gpu_system
+  )
+
+  # 4. Invalid GPU (accelerator type mismatch)
+  invalid_gpu_reservation = _Reservation(
+      name='res2',
+      specificReservation=_SpecificReservation(
+          count=10,
+          inUseCount=0,
+          machine_type='g2-standard-12',
+          guest_accelerators=[
+              _AcceleratorResource(
+                  acceleratorCount=1, acceleratorType='nvidia-wrong'
+              )
+          ],
+      ),
+      aggregateReservation=None,
+  )
+  assert not _verify_reservation_configuration(
+      invalid_gpu_reservation, valid_gpu_system
+  )
+  assert 'ERROR: Reservation' in xpk_print.call_args[0][0]
+
+  # 5. No specific reservation (should pass)
+  no_specific_reservation = _Reservation(
+      name='res3', specificReservation=None, aggregateReservation=None
+  )
+  assert _verify_reservation_configuration(
+      no_specific_reservation, valid_tpu_system
+  )
+
+
+def test_get_reservation_cached_caching(mocker):
+  mock_run_command = mocker.patch(
+      'xpk.core.capacity.run_command_for_value',
+      return_value=(0, '{"name": "res", "status": "READY"}'),
+  )
+
+  # First call
+  _get_reservation_cached(ReservationLink('project', 'res1', 'zone'))
+  assert mock_run_command.call_count == 1
+
+  # Second call with same args
+  _get_reservation_cached(ReservationLink('project', 'res1', 'zone'))
+  assert mock_run_command.call_count == 1  # Should still be 1
+
+  # Third call with different args
+  _get_reservation_cached(ReservationLink('project', 'res2', 'zone'))
+  assert mock_run_command.call_count == 2
+
+
+def test_assess_available_slices_filters_invalid_block_reservation(
+    commands_tester: CommandsTester,
+    test_system: SystemCharacteristics,
+):
+  # Mock the parent reservation being fetched, but with invalid machine type
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 100, "inUseCount": 0,'
+              ' "instanceProperties": {"machineType": "wrong-machine"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      'gcloud beta compute reservations describe',
+  )
+
+  res_block = BlockReservationLink(
+      project='project',
+      name='reservation',
+      zone='zone',
+      block_name='block',
+  )
+
+  slices, return_code = assess_available_slices(
+      [res_block],
+      force_sub_block_targeting=True,
+      required_hosts=1,
+      system=test_system,
+  )
+
+  # Should return empty because validation failed
+  assert not slices
+  assert return_code == 0
+
+
+def test_parse_reservation_with_new_fields():
+  data = {
+      'specificReservation': {
+          'count': '10',
+          'inUseCount': '2',
+          'instanceProperties': {
+              'machineType': 'test-machine',
+              'maintenanceInterval': 'PERIODIC',
+          },
+      },
+      'deploymentType': 'DENSE',
+      'resourcePolicies': {'policy': 'compact-policy'},
+      'status': 'READY',
+  }
+
+  res = _parse_reservation('res1', data)
+
+  assert res.name == 'res1'
+  assert res.deployment_type == 'DENSE'
+  assert res.resource_policy == 'compact-policy'
+  assert res.specificReservation.maintenance_interval == 'PERIODIC'
+
+
+def test_parse_reservation_defaults_new_fields():
+  data = {
+      'specificReservation': {
+          'count': '10',
+          'inUseCount': '2',
+          'instanceProperties': {
+              'machineType': 'test-machine',
+          },
+      },
+      'status': 'READY',
+  }
+
+  res = _parse_reservation('res1', data)
+
+  assert res.name == 'res1'
+  assert res.deployment_type == ''
+  assert res.resource_policy == ''
+  assert res.specificReservation.maintenance_interval == ''
diff --git a/src/xpk/core/cluster.py b/src/xpk/core/cluster.py
index 6f17ab287..3cd30b076 100644
--- a/src/xpk/core/cluster.py
+++ b/src/xpk/core/cluster.py
@@ -15,8 +15,6 @@
 """
 
 import yaml
-from google.api_core.exceptions import PermissionDenied
-from google.cloud import resourcemanager_v3
 from kubernetes import client as k8s_client
 from kubernetes import config
 from kubernetes.client.exceptions import ApiException
@@ -33,6 +31,7 @@
     add_zone_and_project,
     get_cluster_location,
     zone_to_region,
+    project_id_to_project_number,
 )
 from .nodepool import recreate_nodes_in_existing_node_pools
 from .resources import get_cluster_system_characteristics
@@ -389,23 +388,6 @@ def get_all_clusters_programmatic(args) -> tuple[list[str], int]:
   return raw_cluster_output.splitlines(), 0
 
 
-def project_id_to_project_number(project_id: str) -> str:
-  client = resourcemanager_v3.ProjectsClient()
-  request = resourcemanager_v3.GetProjectRequest()
-  request.name = f'projects/{project_id}'
-  try:
-    response = client.get_project(request=request)
-  except PermissionDenied as e:
-    xpk_print(
-        f"Couldn't translate project id: {project_id} to project number."
-        f' Error: {e}'
-    )
-    xpk_exit(1)
-  parts = response.name.split('/', 1)
-  xpk_print(f'Project number for project: {project_id} is {parts[1]}')
-  return str(parts[1])
-
-
 def setup_k8s_env(args) -> k8s_client.ApiClient:
   add_zone_and_project(args)
   get_cluster_credentials(args)
diff --git a/src/xpk/core/gcloud_context.py b/src/xpk/core/gcloud_context.py
index 0bd52abd5..535cee7c5 100644
--- a/src/xpk/core/gcloud_context.py
+++ b/src/xpk/core/gcloud_context.py
@@ -18,6 +18,9 @@
 import sys
 from dataclasses import dataclass
 
+from google.api_core.exceptions import PermissionDenied
+from google.cloud import resourcemanager_v3
+
 from ..utils.console import xpk_print, xpk_exit
 from ..utils.versions import ReleaseChannel
 from .commands import run_command_for_value
@@ -208,8 +211,8 @@ def get_gke_control_plane_version(
 
   if not is_valid_version:
     xpk_print(
-        f'Planned GKE Version: {master_gke_version}\n Valid Versions:'
-        f'\n{gke_server_config.valid_versions}\nRecommended / Default GKE'
+        f'Planned GKE Version: {master_gke_version}\\n Valid Versions:'
+        f'\\n{gke_server_config.valid_versions}\\nRecommended / Default GKE'
         f' Version: {gke_server_config.default_gke_version}'
     )
     xpk_print(
@@ -224,3 +227,21 @@ def get_gke_control_plane_version(
     return 1, None
 
   return 0, master_gke_version
+
+
+@lru_cache()
+def project_id_to_project_number(project_id: str) -> str:
+  client = resourcemanager_v3.ProjectsClient()
+  request = resourcemanager_v3.GetProjectRequest()
+  request.name = f'projects/{project_id}'
+  try:
+    response = client.get_project(request=request)
+  except PermissionDenied as e:
+    xpk_print(
+        f"Couldn't translate project id: {project_id} to project number."
+        f' Error: {e}'
+    )
+    xpk_exit(1)
+  parts = response.name.split('/', 1)
+  xpk_print(f'Project number for project: {project_id} is {parts[1]}')
+  return str(parts[1])
diff --git a/src/xpk/core/nodepool.py b/src/xpk/core/nodepool.py
index 490829961..c07c9bd20 100644
--- a/src/xpk/core/nodepool.py
+++ b/src/xpk/core/nodepool.py
@@ -15,7 +15,7 @@
 """
 
 from typing import Iterator, List
-from itertools import cycle
+from itertools import chain, repeat
 
 from ..utils.console import ask_for_user_consent, xpk_print
 from .scheduling import get_placement_policy_name, is_placement_policy_supported
@@ -27,6 +27,7 @@
     get_capacity_type,
     get_reservations_list,
     print_reservations,
+    assess_available_slices,
     to_reservation_path,
     ReservationLink,
 )
@@ -254,8 +255,8 @@ def run_gke_node_pool_create_command(
         return 1
 
   placement_args = ''
+  super_slicing: bool = args.super_slicing
   if is_placement_policy_supported(system):
-    super_slicing = args.super_slicing
     placement_policy = get_placement_policy_name(
         system,
         super_slicing,
@@ -277,14 +278,19 @@ def run_gke_node_pool_create_command(
   reservations_iter: Iterator[ReservationLink] | None = None
   if capacity_type == CapacityType.RESERVATION:
     reservations = get_reservations_list(args)
-    if (
-        _validate_reservation_count(reservations, len(node_pools_to_create))
-        != 0
-    ):
-      return 1
-    reservations_iter = (
-        cycle(reservations) if len(reservations) == 1 else iter(reservations)
+    reservations_iter, return_code = _prepare_reservation_iterator(
+        reservations=reservations,
+        num_new_node_pools=len(node_pools_to_create),
+        force_sub_block_targeting=super_slicing,
+        required_hosts=(
+            max_nodes
+            if system.accelerator_type == AcceleratorType.TPU
+            else args.num_nodes
+        ),
+        system=system,
     )
+    if return_code > 0:
+      return return_code
 
   for node_pool_name in node_pools_to_create:
     capacity_args, return_code = get_capacity_arguments_from_capacity_type(
@@ -684,19 +690,6 @@ def ensure_resource_policy_exists(
     raise RuntimeError('Unable to create resource policy')
 
 
-def _validate_reservation_count(
-    reservations: List[ReservationLink], num_node_pools_to_create: int
-) -> int:
-  """Validate that reservation count matches new nodepool count or is 1."""
-  if len(reservations) > 1 and len(reservations) != num_node_pools_to_create:
-    xpk_print(
-        f'Error: Number of reservations ({len(reservations)}) must match'
-        f' the number of NEW nodepools ({num_node_pools_to_create}) or be 1.'
-    )
-    return 1
-  return 0
-
-
 def recreate_nodes_in_existing_node_pools(args) -> int:
   """Triggers a manual upgrade of nodepools to the same version to force recreation
   of nodes.
@@ -729,3 +722,39 @@ def recreate_nodes_in_existing_node_pools(args) -> int:
       task_names,
   )
   return maybe_failure.return_code if maybe_failure is not None else 0
+
+
+def _prepare_reservation_iterator(
+    reservations: List[ReservationLink],
+    num_new_node_pools: int,
+    force_sub_block_targeting: bool,
+    required_hosts: int,
+    system: SystemCharacteristics,
+) -> tuple[Iterator[ReservationLink] | None, int]:
+  available_capacity, return_code = assess_available_slices(
+      reservations,
+      force_sub_block_targeting=force_sub_block_targeting,
+      required_hosts=required_hosts,
+      system=system,
+  )
+
+  if return_code > 0:
+    return None, return_code
+
+  total_available = sum(cap.available_slices for cap in available_capacity)
+
+  if total_available < num_new_node_pools:
+    xpk_print(
+        'Error: Not enough available reservation capacity. Needed'
+        f' {num_new_node_pools} slices, but only found'
+        f' {total_available} healthy and fitting in the provided'
+        ' reservations.'
+    )
+    return None, 1
+
+  reservations_iter = chain.from_iterable(
+      repeat(cap.reservation, cap.available_slices)
+      for cap in available_capacity
+  )
+
+  return reservations_iter, 0
diff --git a/src/xpk/core/nodepool_test.py b/src/xpk/core/nodepool_test.py
index d6ba5d341..a6d93ff5e 100644
--- a/src/xpk/core/nodepool_test.py
+++ b/src/xpk/core/nodepool_test.py
@@ -21,12 +21,10 @@
     get_desired_node_pool_names,
     run_gke_node_pool_create_command,
     recreate_nodes_in_existing_node_pools,
-    _validate_reservation_count,
 )
 from xpk.core.system_characteristics import AcceleratorType, SystemCharacteristics, DockerPlatform, GpuConfig
 from xpk.core.commands import FailedCommand
 from xpk.core.testing.commands_tester import CommandsTester
-from xpk.core.capacity import ReservationLink
 
 
 CLUSTER_NAME = "running-cucumber"
@@ -436,23 +434,6 @@ def test_display_nodepool_creation_ignores_logs_without_errors(
   )
 
 
-def test_validate_reservation_count_mismatch(mock_xpk_print):
-  result = _validate_reservation_count(
-      [
-          ReservationLink(project="p", name="res1", zone="z"),
-          ReservationLink(project="p", name="res2", zone="z"),
-      ],
-      num_node_pools_to_create=3,
-  )
-
-  assert result == 1
-  assert mock_xpk_print.call_count == 1
-  assert (
-      "reservations (2) must match the number of NEW nodepools (3)"
-      in mock_xpk_print.call_args_list[0].args[0]
-  )
-
-
 def test_run_gke_node_pool_create_command_multiple_reservations(
     mocker,
     commands_tester: CommandsTester,
@@ -463,7 +444,7 @@ def test_run_gke_node_pool_create_command_multiple_reservations(
   mocker.patch("xpk.core.capacity.verify_reservations_exist", return_value=0)
   args = mocker.Mock(
       num_slices=2,
-      reservation="res1,res2",
+      reservation="reservation1,reservation2",
       tpu_type="v4-8",
       device_type=None,
       cluster="test-cluster",
@@ -476,6 +457,7 @@ def test_run_gke_node_pool_create_command_multiple_reservations(
       enable_gcsfuse_csi_driver=False,
       host_maintenance_interval="AS_NEEDED",
       custom_nodepool_arguments="",
+      super_slicing=False,
   )
   system = SystemCharacteristics(
       topology="2x2x1",
@@ -494,7 +476,17 @@ def test_run_gke_node_pool_create_command_multiple_reservations(
   commands_tester.set_result_for_command(
       (0, ""), "gcloud beta container node-pools list"
   )
-
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 2, "inUseCount": 0,'
+              ' "instanceProperties": {"machineType": "ct4p-hightpu-4t"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      "gcloud beta compute reservations describe",
+  )
   result = run_gke_node_pool_create_command(args, system, "1.2.3")
 
   assert result == 0
@@ -502,10 +494,16 @@ def test_run_gke_node_pool_create_command_multiple_reservations(
       "gcloud", "node-pools create", "--tpu-topology=2x2x1", times=2
   )
   commands_tester.assert_command_run(
-      "gcloud", "node-pools create", "test-cluster-np-0", "--reservation=res1"
+      "gcloud",
+      "node-pools create",
+      "test-cluster-np-0",
+      "--reservation=reservation1",
   )
   commands_tester.assert_command_run(
-      "gcloud", "node-pools create", "test-cluster-np-1", "--reservation=res2"
+      "gcloud",
+      "node-pools create",
+      "test-cluster-np-1",
+      "--reservation=reservation2",
   )
 
 
@@ -520,7 +518,7 @@ def test_run_gke_node_pool_create_command_partial_reservations(
   mocker.patch("xpk.core.capacity.verify_reservations_exist", return_value=0)
   args = mocker.Mock(
       num_slices=3,
-      reservation="res1,res2",
+      reservation="reservation1,reservation2",
       tpu_type="v4-8",
       device_type=None,
       cluster="test-cluster",
@@ -533,6 +531,7 @@ def test_run_gke_node_pool_create_command_partial_reservations(
       enable_gcsfuse_csi_driver=False,
       host_maintenance_interval="AS_NEEDED",
       custom_nodepool_arguments="",
+      super_slicing=False,
   )
   system = SystemCharacteristics(
       topology="2x2x1",
@@ -557,7 +556,17 @@ def test_run_gke_node_pool_create_command_partial_reservations(
       "node-pools describe",
       '--format="value(locations)"',
   )
-
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '{"specificReservation": {"count": 2, "inUseCount": 0,'
+              ' "instanceProperties": {"machineType": "ct4p-hightpu-4t"}},'
+              ' "status": "READY"}'
+          ),
+      ),
+      "gcloud beta compute reservations describe",
+  )
   result = run_gke_node_pool_create_command(args, system, "1.2.3")
 
   assert result == 0
@@ -565,10 +574,16 @@ def test_run_gke_node_pool_create_command_partial_reservations(
       "gcloud", "node-pools create", "--tpu-topology=2x2x1", times=2
   )
   commands_tester.assert_command_run(
-      "gcloud", "node-pools create", "test-cluster-np-1", "--reservation=res1"
+      "gcloud",
+      "node-pools create",
+      "test-cluster-np-1",
+      "--reservation=reservation1",
   )
   commands_tester.assert_command_run(
-      "gcloud", "node-pools create", "test-cluster-np-2", "--reservation=res2"
+      "gcloud",
+      "node-pools create",
+      "test-cluster-np-2",
+      "--reservation=reservation2",
   )
 
 
@@ -588,6 +603,7 @@ def test_recreate_nodes_in_existing_node_pools_upgrades_existing_nodepools(
       project="test-project",
       zone="us-central1-a",
   )
+
   commands_tester.set_result_for_command(
       (0, ""), "gcloud container clusters upgrade"
   )
@@ -672,3 +688,136 @@ def test_recreate_nodes_in_existing_node_pools_returns_error_code_if_upgrade_fai
       "--location=us-central1",
       "--quiet",
   )
+
+
+def test_run_gke_node_pool_create_command_super_slicing_exhaustion(
+    mocker,
+    commands_tester: CommandsTester,
+):
+  mocker.patch(
+      "xpk.core.nodepool.get_cluster_location", return_value="us-central1"
+  )
+  mocker.patch("xpk.core.capacity.verify_reservations_exist", return_value=0)
+  args = mocker.Mock(
+      num_slices=2,
+      reservation="reservation1",
+      tpu_type="v4-8",
+      device_type=None,
+      cluster="test-cluster",
+      project="test-project",
+      zone="us-central1-a",
+      on_demand=False,
+      spot=False,
+      flex=False,
+      enable_workload_identity=False,
+      enable_gcsfuse_csi_driver=False,
+      host_maintenance_interval="AS_NEEDED",
+      custom_nodepool_arguments="",
+      super_slicing=True,
+      num_nodes=2,
+  )
+  system = SystemCharacteristics(
+      topology="2x2x1",
+      vms_per_slice=2,
+      gke_accelerator="tpu-v4",
+      gce_machine_type="ct4p-hightpu-4t",
+      chips_per_vm=4,
+      accelerator_type=AcceleratorType.TPU,
+      device_type="v4-8",
+      requires_workload_policy=False,
+      supports_sub_slicing=False,
+      supports_super_slicing=False,
+      supports_accelerator_network_profile=True,
+      docker_platform=DockerPlatform.AMD,
+  )
+  commands_tester.set_result_for_command(
+      (
+          0,
+          (
+              '[{"name": "sub-block1", "count": 2, "inUseCount": 0}, {"name":'
+              ' "sub-block2", "count": 2, "inUseCount": 0}]'
+          ),
+      ),
+      "gcloud beta compute reservations sub-blocks list",
+  )
+  commands_tester.set_result_for_command(
+      (0, "block1"), "gcloud beta compute reservations blocks list"
+  )
+
+  result = run_gke_node_pool_create_command(args, system, "1.2.3")
+
+  assert result == 0
+  commands_tester.assert_command_run(
+      "gcloud",
+      "node-pools create",
+      "test-cluster-np-0",
+      "--reservation=reservation1/reservationBlocks/block1/reservationSubBlocks/sub-block1",
+  )
+  commands_tester.assert_command_run(
+      "gcloud",
+      "node-pools create",
+      "test-cluster-np-1",
+      "--reservation=reservation1/reservationBlocks/block1/reservationSubBlocks/sub-block2",
+  )
+
+
+def test_run_gke_node_pool_create_command_super_slicing_insufficient_capacity(
+    mocker,
+    commands_tester: CommandsTester,
+    mock_xpk_print,
+):
+  """Requested 2 slices with vms_per_slice=2, but there's only 1 slice available."""
+  mocker.patch(
+      "xpk.core.nodepool.get_cluster_location", return_value="us-central1"
+  )
+  mocker.patch("xpk.core.capacity.verify_reservations_exist", return_value=0)
+  args = mocker.Mock(
+      num_slices=2,
+      reservation="reservation1/reservationBlocks/block1",
+      tpu_type="v4-8",
+      device_type=None,
+      cluster="test-cluster",
+      project="test-project",
+      zone="us-central1-a",
+      on_demand=False,
+      spot=False,
+      flex=False,
+      enable_workload_identity=False,
+      enable_gcsfuse_csi_driver=False,
+      host_maintenance_interval="AS_NEEDED",
+      custom_nodepool_arguments="",
+      super_slicing=True,
+      num_nodes=2,
+  )
+  system = SystemCharacteristics(
+      topology="2x2x1",
+      vms_per_slice=2,
+      gke_accelerator="tpu-v4",
+      gce_machine_type="ct4p-hightpu-4t",
+      chips_per_vm=4,
+      accelerator_type=AcceleratorType.TPU,
+      device_type="v4-8",
+      requires_workload_policy=False,
+      supports_sub_slicing=False,
+      supports_super_slicing=False,
+      supports_accelerator_network_profile=True,
+      docker_platform=DockerPlatform.AMD,
+  )
+  commands_tester.set_result_for_command(
+      (0, "block1"), "gcloud beta compute reservations blocks list"
+  )
+  commands_tester.set_result_for_command(
+      (
+          0,
+          '[{"name": "sub-block1", "count": 2, "inUseCount": 0}]',
+      ),
+      "gcloud beta compute reservations sub-blocks list",
+  )
+  result = run_gke_node_pool_create_command(args, system, "1.2.3")
+
+  assert result == 1
+  assert mock_xpk_print.call_count >= 1
+  assert (
+      "Not enough available reservation capacity"
+      in mock_xpk_print.call_args_list[-1].args[0]
+  )
diff --git a/src/xpk/core/system_characteristics.py b/src/xpk/core/system_characteristics.py
index 47c67e8b6..3323791f4 100644
--- a/src/xpk/core/system_characteristics.py
+++ b/src/xpk/core/system_characteristics.py
@@ -132,6 +132,7 @@ class SystemCharacteristics:
     requires_workload_policy: A boolean indicating if a GCE resource
       workload policy is required. This is automatically set to True for GPUs.
     parallel_containers: The number of containers running on a single VM.
+    reservation_accelerator_type: The accelerator type used for reservation.
 
   """
 
@@ -168,6 +169,14 @@ def gpu_requires_topology(self) -> bool:
     """
     return self.gpu_config.requires_topology if self.gpu_config else False
 
+  @property
+  def reservation_accelerator_type(self) -> str | None:
+    if self.accelerator_type == AcceleratorType.TPU:
+      return self.gce_machine_type.split('-')[0]
+    elif self.accelerator_type == AcceleratorType.GPU:
+      return self.gke_accelerator
+    return None
+
 
 def get_system_characteristics(
     args,
@@ -312,8 +321,8 @@ def compute_vms_per_slice(topology: str) -> int:
         supports_sub_slicing=False,
         supports_super_slicing=False,
         supports_accelerator_network_profile=False,
-        gpu_config=GpuConfig(requires_topology=False),
         docker_platform=AMD_PLATFORM,
+        gpu_config=GpuConfig(requires_topology=False),
     ),
     'l4-2': SystemCharacteristics(
         topology='N/A',
@@ -326,8 +335,8 @@ def compute_vms_per_slice(topology: str) -> int:
         supports_sub_slicing=False,
         supports_super_slicing=False,
         supports_accelerator_network_profile=False,
-        gpu_config=GpuConfig(requires_topology=False),
         docker_platform=AMD_PLATFORM,
+        gpu_config=GpuConfig(requires_topology=False),
     ),
     'l4-4': SystemCharacteristics(
         topology='N/A',
@@ -340,8 +349,8 @@ def compute_vms_per_slice(topology: str) -> int:
         supports_sub_slicing=False,
         supports_super_slicing=False,
         supports_accelerator_network_profile=False,
-        gpu_config=GpuConfig(requires_topology=False),
         docker_platform=AMD_PLATFORM,
+        gpu_config=GpuConfig(requires_topology=False),
     ),
     'l4-8': SystemCharacteristics(
         topology='N/A',
@@ -354,8 +363,8 @@ def compute_vms_per_slice(topology: str) -> int:
         supports_sub_slicing=False,
         supports_super_slicing=False,
         supports_accelerator_network_profile=False,
-        gpu_config=GpuConfig(requires_topology=False),
         docker_platform=AMD_PLATFORM,
+        gpu_config=GpuConfig(requires_topology=False),
     ),
     # A100-40gb-$CHIPSc
     'a100-40gb-1': SystemCharacteristics(
@@ -369,8 +378,8 @@ def compute_vms_per_slice(topology: str) -> int:
         supports_sub_slicing=False,
         supports_super_slicing=False,
         supports_accelerator_network_profile=False,
-        gpu_config=GpuConfig(requires_topology=False),
         docker_platform=AMD_PLATFORM,
+        gpu_config=GpuConfig(requires_topology=False),
     ),
     'a100-40gb-2': SystemCharacteristics(
         topology='N/A',
@@ -383,8 +392,8 @@ def compute_vms_per_slice(topology: str) -> int:
         supports_sub_slicing=False,
         supports_super_slicing=False,
         supports_accelerator_network_profile=False,
-        gpu_config=GpuConfig(requires_topology=False),
         docker_platform=AMD_PLATFORM,
+        gpu_config=GpuConfig(requires_topology=False),
     ),
     'a100-40gb-4': SystemCharacteristics(
         topology='N/A',
@@ -397,8 +406,8 @@ def compute_vms_per_slice(topology: str) -> int:
         supports_sub_slicing=False,
         supports_super_slicing=False,
         supports_accelerator_network_profile=False,
-        gpu_config=GpuConfig(requires_topology=False),
         docker_platform=AMD_PLATFORM,
+        gpu_config=GpuConfig(requires_topology=False),
     ),
     'a100-40gb-8': SystemCharacteristics(
         topology='N/A',
@@ -411,8 +420,8 @@ def compute_vms_per_slice(topology: str) -> int:
         supports_sub_slicing=False,
         supports_super_slicing=False,
         supports_accelerator_network_profile=False,
-        gpu_config=GpuConfig(requires_topology=False),
         docker_platform=AMD_PLATFORM,
+        gpu_config=GpuConfig(requires_topology=False),
     ),
     'gb200-4': SystemCharacteristics(
         topology='1x72',
@@ -425,13 +434,13 @@ def compute_vms_per_slice(topology: str) -> int:
         supports_sub_slicing=False,
         supports_super_slicing=False,
         supports_accelerator_network_profile=True,
+        docker_platform=ARM_PLATFORM,
         gpu_config=GpuConfig(
             requires_topology=True,
             nccl_installer=INSTALLER_NCCL_RDMA_A4X,
             jobset_decorator_fn=rdma_decorator.decorate_jobset,
             gpu_direct_name='rdma',
         ),
-        docker_platform=ARM_PLATFORM,
     ),
     'gb200-4-nolssd': SystemCharacteristics(
         topology='1x72',
@@ -444,13 +453,13 @@ def compute_vms_per_slice(topology: str) -> int:
         supports_sub_slicing=False,
         supports_super_slicing=False,
         supports_accelerator_network_profile=True,
+        docker_platform=ARM_PLATFORM,
         gpu_config=GpuConfig(
             requires_topology=True,
             nccl_installer=INSTALLER_NCCL_RDMA_A4X,
             jobset_decorator_fn=rdma_decorator.decorate_jobset,
             gpu_direct_name='rdma',
         ),
-        docker_platform=ARM_PLATFORM,
     ),
     'b200-8': SystemCharacteristics(
         topology='N/A',
@@ -463,13 +472,13 @@ def compute_vms_per_slice(topology: str) -> int:
         supports_sub_slicing=False,
         supports_super_slicing=False,
         supports_accelerator_network_profile=True,
+        docker_platform=AMD_PLATFORM,
         gpu_config=GpuConfig(
             requires_topology=True,
             nccl_installer=INSTALLER_NCCL_RDMA,
             jobset_decorator_fn=rdma_decorator.decorate_jobset,
             gpu_direct_name='rdma',
         ),
-        docker_platform=AMD_PLATFORM,
     ),
     'h200-141gb-8': SystemCharacteristics(
         topology='N/A',
@@ -482,13 +491,13 @@ def compute_vms_per_slice(topology: str) -> int:
         supports_sub_slicing=False,
         supports_super_slicing=False,
         supports_accelerator_network_profile=True,
+        docker_platform=AMD_PLATFORM,
         gpu_config=GpuConfig(
             requires_topology=True,
             nccl_installer=INSTALLER_NCCL_RDMA,
             jobset_decorator_fn=rdma_decorator.decorate_jobset,
             gpu_direct_name='rdma',
         ),
-        docker_platform=AMD_PLATFORM,
     ),
     # H100-80gb-$CHIPS
     'h100-80gb-8': SystemCharacteristics(
@@ -502,13 +511,13 @@ def compute_vms_per_slice(topology: str) -> int:
         supports_sub_slicing=False,
         supports_super_slicing=False,
         supports_accelerator_network_profile=True,
+        docker_platform=AMD_PLATFORM,
         gpu_config=GpuConfig(
             requires_topology=True,
             nccl_installer=INSTALLER_NCCL_TCPX,
             jobset_decorator_fn=tcpx_decorator.decorate_jobset,
             gpu_direct_name='tcpx',
         ),
-        docker_platform=AMD_PLATFORM,
     ),
     # H100-mega-80gb-$CHIPS
     'h100-mega-80gb-8': SystemCharacteristics(
@@ -522,13 +531,13 @@ def compute_vms_per_slice(topology: str) -> int:
         supports_sub_slicing=False,
         supports_super_slicing=False,
         supports_accelerator_network_profile=True,
+        docker_platform=AMD_PLATFORM,
         gpu_config=GpuConfig(
             requires_topology=True,
             nccl_installer=INSTALLER_NCCL_TCPXO,
             jobset_decorator_fn=tcpxo_decorator.decorate_jobset,
             gpu_direct_name='tcpxo',
         ),
-        docker_platform=AMD_PLATFORM,
     ),
     # TPU system characteristics
     **get_tpu_system_characteristics_map(
diff --git a/src/xpk/core/system_characteristics_test.py b/src/xpk/core/system_characteristics_test.py
index 1dfd9905e..c1c9fe232 100644
--- a/src/xpk/core/system_characteristics_test.py
+++ b/src/xpk/core/system_characteristics_test.py
@@ -238,3 +238,52 @@ def test_system_characteristics_post_init_throws_for_gpu_without_config():
         supports_accelerator_network_profile=False,
         docker_platform=DockerPlatform.AMD,
     )
+
+
+def test_reservation_accelerator_type_derived_correctly():
+  """Tests that reservation_accelerator_type is correctly derived."""
+  tpu_system = SystemCharacteristics(
+      topology="2x2x1",
+      vms_per_slice=1,
+      gke_accelerator="tpu-v5p-slice",
+      gce_machine_type="ct5p-hightpu-4t",
+      chips_per_vm=4,
+      accelerator_type=AcceleratorType.TPU,
+      device_type="v5p-8",
+      supports_sub_slicing=False,
+      supports_super_slicing=False,
+      supports_accelerator_network_profile=False,
+      docker_platform=DockerPlatform.AMD,
+  )
+  assert tpu_system.reservation_accelerator_type == "ct5p"
+
+  gpu_system = SystemCharacteristics(
+      topology="N/A",
+      vms_per_slice=1,
+      gke_accelerator="nvidia-l4",
+      gce_machine_type="g2-standard-12",
+      chips_per_vm=1,
+      accelerator_type=AcceleratorType.GPU,
+      device_type="l4-1",
+      supports_sub_slicing=False,
+      supports_super_slicing=False,
+      supports_accelerator_network_profile=False,
+      docker_platform=DockerPlatform.AMD,
+      gpu_config=GpuConfig(requires_topology=False),
+  )
+  assert gpu_system.reservation_accelerator_type == "nvidia-l4"
+
+  cpu_system = SystemCharacteristics(
+      topology="N/A",
+      vms_per_slice=1,
+      gke_accelerator="N/A",
+      gce_machine_type="n2-standard-32",
+      chips_per_vm=32,
+      accelerator_type=AcceleratorType.CPU,
+      device_type="n2-standard-32-1",
+      supports_sub_slicing=False,
+      supports_super_slicing=False,
+      supports_accelerator_network_profile=False,
+      docker_platform=DockerPlatform.AMD,
+  )
+  assert cpu_system.reservation_accelerator_type is None