Alternative API changelog - 2.2 Release | hosted·ai Documentation

Previous Release: 2.1 Release New Release: 2.2

This document covers API changes that affect customers with existing integrations for billing and GPUaaS instance creation. Please review all sections marked BREAKING carefully and update your integrations before upgrading.

1. Breaking Changes Summary

The following changes require updates to existing API integrations:

Change	Impact	Action Required
Instance creation request body restructured	All instance creation calls will fail	Update request payloads to new format
GPUaaS pod creation moved from subscription to instance API	`POST /gpuaas/pool/subscribe` no longer used for pod creation	Use `POST /service/i/create-instance` with `pod_opts` and a `pod_accelerator` service
GPUaaS pod costs move to team/workspace billing	`/team-billing/gpuaas-pool/` returns empty for migrated subs; historical data moves	Use team/workspace billing with `nature: "pod"` filter
Billing Swagger documentation corrected	Swagger/OpenAPI spec consumers may need updates	See note — actual API behavior unchanged
Billing response format changed	Billing parsers will break	Update response parsing (PCI devices moved per-instance, new metadata fields)
Instance credentials response changed	Credential parsing may break	Handle nullable fields, new `ip` and `port` fields
`/service/i/details` removed	Service detail calls will return 404	Use `/service/i/provisioning-info` instead
`PoolWorkersConnectionDetails` fields now nullable	Worker connection parsing may break	Handle nullable `username`, `password`, `port`, `ip`
New resource fields use float types	Clients using old integer-typed resource structs	Use new float-typed response fields
Rate limiting added (60 req/min)	High-frequency clients may get 429 errors	Implement rate limit handling and backoff
`GET /gpuaas/sub/{sub_id}/pool/{pool_id}/target-url` removed	Ray head port calls will fail	Remove calls to this endpoint

2. Instance Creation API Changes

BREAKING: New Request Body Format

The instance creation endpoint POST /service/i/create-instance now uses a unified request body that supports both VM and Pod instance types.

Old Request Format:

{ "service_id": "uuid", "name": "my-instance", "description": "string", "workspace_id": "string", "team_id": "uuid", "region_id": 1, "cluster_id": 1, "image_id": 1, "instance_type_id": "string", "root_storage_type_id": "string", "additional_disk_one_type_id": "string", "additional_disk_two_type_id": "string", "networks": [1, 2], "public_keys": ["ssh-rsa ..."], "pool_ids": [1], "pcidev_ids": [1], "gpu_card_count": 1, "gpu_model": "string", "recipe_id": 1 }

New Request Format:

{ "service_id": "uuid", "name": "my-instance", "description": "string", "workspace_id": "uuid", "team_id": "uuid", "region_id": 1, "image_hash": "string", "instance_type_id": "uuid", "root_storage_type_id": "uuid", "public_keys": ["ssh-rsa ..."], "vm_opts": { "additional_disk_one_type_id": "string", "additional_disk_two_type_id": "string", "networks": [1, 2], "passthrough_accelerators": "string", "gpu_card_count": 1 }, "pod_opts": { "pool_id": 1, "vgpus": 4, "shared_volumes": [1, 2] } }

Key field changes:

Field	Old	New	Notes
`service_id`	uuid (optional, pointer)	uuid (required, value)	Now required — must always be provided
`image_id`	integer	Removed	Replaced by `image_hash` (string)
`image_hash`	N/A	string	New field — images are now referenced by hash
`cluster_id`	integer	Removed	No longer needed; region is derived from cluster
`region_id`	int	uint	Type changed from signed to unsigned integer
`workspace_id`	string	uuid	Type changed
`instance_type_id`	string	*uuid (optional)	Type changed to nullable UUID pointer — omit or set null if not applicable
`root_storage_type_id`	string	*uuid (optional)	Type changed to nullable UUID pointer — omit or set null if not applicable
`pool_ids`	integer array	Removed	Replaced by `pod_opts.pool_id` (single uint, not an array)
`pcidev_ids`	integer array	Removed	Removed entirely — no direct replacement
`gpu_model`	string	Removed	Replaced by `vm_opts.passthrough_accelerators` (string)
`recipe_id`	integer (optional)	Removed	Managed internally — no longer a client-supplied field
`vm_opts`	N/A	object (optional)	New — VM-specific options grouped here
`pod_opts`	N/A	object (optional)	New — Pod/GPUaaS-specific options grouped here

Important: For VM services, provide vm_opts. For Pod/GPUaaS services (service type pod_accelerator), provide pod_opts. Providing the wrong options object for the service type will return an error.

New Instance Response Fields

Instance responses now include:

Field	Type	Description
`nature`	enum: `vm`, `pod`	Whether the instance is a VM or a Pod
`vm_public_ips`	string array	Public IP addresses assigned to the instance

3. GPUaaS Pod Creation: Subscription to Instance Migration

BREAKING: Pods Are Now Created as Instances, Not Subscriptions

In the previous release, GPUaaS pods were created by subscribing to a GPU pool via a dedicated endpoint. In Titan, GPUaaS pods are now first-class instances created through the same unified instance creation endpoint as VMs.

Old Approach (Oct Release): Pool Subscription

Endpoint: POST /gpuaas/pool/subscribe

Request body (SubscribeToPoolPayload):

{ "pool_id": 1, "team_id": "uuid", "is_vip_priority": false, "vgpus": 4, "guaranteed_gpu_share_percent": 100, "instance_type_id": "uuid", "ephemeral_storage_block_id": "uuid", "shared_volumes": [1, 2], "image_uuid": "uuid", "port_guid": "string (optional)", "ifb_subscription_id": "string (optional)" }

This created a PoolSubscription record, and a pod was provisioned asynchronously as a side effect of the subscription. The pod did not appear in the instances list and was managed entirely through the GPUaaS subscription system.

New Approach (Titan): Unified Instance Creation

Endpoint: POST /service/i/create-instance

Request body (same endpoint as VM creation, using pod_opts):

{ "service_id": "uuid (must be a pod_accelerator service)", "name": "my-pod", "description": "string", "workspace_id": "uuid", "team_id": "uuid", "region_id": 1, "image_hash": "string", "instance_type_id": "uuid", "root_storage_type_id": "uuid", "public_keys": ["ssh-rsa ..."], "pod_opts": { "pool_id": 1, "vgpus": 4, "shared_volumes": [1, 2] } }

The backend detects the pod_accelerator service type and provisions a pod directly. The pod is created as an Instance record with nature: "pod" and appears in the unified instances list alongside VMs.

Field Mapping: Old Subscription to New Instance Creation

Old Field (`SubscribeToPoolPayload`)	New Location	Notes
`pool_id`	`pod_opts.pool_id`	Same concept, nested under `pod_opts`
`team_id`	`team_id`	Moved to top-level shared field
`vgpus`	`pod_opts.vgpus`	Same concept, nested under `pod_opts`
`instance_type_id`	`instance_type_id`	Moved to top-level shared field
`ephemeral_storage_block_id`	`root_storage_type_id`	Renamed — same concept
`shared_volumes`	`pod_opts.shared_volumes`	Same concept, nested under `pod_opts`
`image_uuid`	`image_hash`	Renamed — images now referenced by hash string
`is_vip_priority`	Removed	No longer a user-supplied option
`guaranteed_gpu_share_percent`	Removed	No longer a user-supplied option
`port_guid`	Removed	Infiniband handling is now managed internally
`ifb_subscription_id`	Removed	Infiniband handling is now managed internally
N/A	`service_id`	New required field — must reference a `pod_accelerator` service
N/A	`name`	New required field — human-readable instance name
N/A	`workspace_id`	New required field — instances belong to a workspace
N/A	`region_id`	New required field — explicit region selection
N/A	`public_keys`	New optional field — SSH public keys for pod access

Key Behavioral Differences

Aspect	Old (Subscription)	New (Instance)
Conceptual model	"Subscribe to a GPU pool" — pod is a side effect	"Create an instance" — pod is a first-class resource
Visibility	Pod did not appear in instance listings	Pod appears in `/instances/unified` alongside VMs
Billing	Only via `/team-billing/gpuaas-pool/`	Appears in team/workspace billing with `nature: "pod"`
Lifecycle management	Managed via subscribe/unsubscribe endpoints	Managed via unified instance endpoints (resize, deprovision, volume actions)
GPU allocation	Implicit during provisioning	Explicit pre-reservation via HaiDra before provisioning; auto-cleanup on failure
Service selection	No service concept — user directly subscribed to a pool	Must select a `pod_accelerator` service, which determines available pools, instance types, and images
Workspace assignment	Pods had no workspace	Pods belong to a workspace (required for billing and access control)

Unsubscribe Replaced by Instance Deletion

Old: POST /gpuaas/pool/unsubscribe with a subscription ID.

New: Delete or deprovision the instance through the unified instance management endpoints. The pod deprovisioning flow handles GPU release and cleanup automatically.

4. Billing API Changes

Swagger Documentation Correction: `timezone` and `region_id` Parameters

Note: The previous Swagger/OpenAPI specification incorrectly documented timezone and region_id as required path parameters. In practice, the actual API has always accepted these as optional query parameters. The Titan release corrects the Swagger documentation to match the actual behavior.

If you are calling the API directly, no changes are needed — the endpoint URLs and parameter handling are unchanged.

If you generated client code from the Swagger spec, you may need to regenerate your client, as the spec now correctly reflects these as optional query parameters rather than required path segments.

The billing endpoint URLs remain:

Endpoint	URL	Query Parameters
Team billing	`GET /team-billing/{team_id}/{start_date}/{end_date}/{interval}`	`timezone` (optional, enum: `local`, `utc`, default: `utc`)
Shared storage billing	`GET /team-billing/shared-storage/{team_id}/{start_date}/{end_date}/{interval}`	`region_id` (optional), `timezone` (optional)
GPUaaS pool billing	`GET /team-billing/gpuaas-pool/{team_id}/{start_date}/{end_date}/{interval}`	`region_id` (optional), `timezone` (optional)
Workspace billing	`GET /workspace-billing/{workspace_id}/{start_date}/{end_date}/{interval}`	`timezone` (optional)
Team billing by workspace	`GET /team-billing/group-by-workspace/{team_id}/{start_date}/{end_date}/{interval}`	`timezone` (optional)

BREAKING: Billing Estimate Request

The billing estimate endpoint POST /calculate-instance-billing-estimate now accepts the same new UnifiedInstanceCreateOpts body described in Section 2.

BREAKING: GPUaaS Pod Costs Move to Team/Workspace Billing

In the previous release, GPUaaS pod costs were only visible through the dedicated GPUaaS pool billing endpoint (/team-billing/gpuaas-pool/). In Titan, legacy GPUaaS pods are migrated to unified instances with nature: "pod", and their costs (GPU, vRAM, TFlops, Subscription Rate, Ephemeral Storage, CPU, RAM) now appear in the standard team and workspace billing responses alongside VM instance costs.

Important: Historical billing data moves with the migration. When a legacy GPUaaS pod is migrated to a unified instance:

The new instance preserves the original creation timestamp of the pool subscription, so historical billing queries (e.g., for periods months before the upgrade) will return the migrated pod as an instance with nature: "pod".
The original pool_subscriptions row is permanently deleted from the database during migration. This means the dedicated GPUaaS pool billing endpoint (/team-billing/gpuaas-pool/) will no longer return data for migrated subscriptions — including historical periods.
There is no double-counting: costs appear exclusively in the team/workspace billing view after migration.

If you rely on the /team-billing/gpuaas-pool/ endpoint:

After upgrade, this endpoint will return empty or reduced results as migrated subscriptions are removed from the underlying data.
To retrieve GPUaaS pod costs, use the standard team or workspace billing endpoints and filter by nature: "pod".
The /team-billing/gpuaas-pool/ endpoint remains functional for any non-migrated pool subscriptions, but all legacy pods are migrated during the upgrade process.

BREAKING: Billing Response Format Changes

PCI device costs consolidated per-instance:

The top-level pci_devices key has been removed from team and workspace billing responses.
PCI device costs are now included within each instance's billing data under a pci_dev key.

New per-instance metadata fields in billing responses:

Team and workspace billing responses now include both VM and Pod instances. Each instance entry includes new metadata fields:

Field	Type	Description
`instance_name`	string	Human-readable instance name
`nature`	enum: `vm`, `pod`	Whether this is a VM or Pod instance
`model_type`	string (optional)	GPU model type (pod instances only)
`pool_id`	integer (optional)	GPUaaS pool ID (pod instances only)
`pool_name`	string (optional)	GPUaaS pool name (pod instances only)
`sharing_ratio`	uint (optional)	GPU sharing ratio (pod instances only)
`vendor_type`	string (optional)	GPU vendor type (pod instances only)
`instance_type`	object (optional)	Instance type details including ephemeral storage (pod instances only)
`total_cost`	decimal (optional)	Total cost for the billing period (pod instances only)

Pod instance billing resources: For instances with nature: "pod", the Resources map contains pod-specific cost categories: Subscription Rate, GPU, vRAM, CPU, RAM, Ephemeral Storage, and TFlops — rather than the VM-style resources (Disk, IP, etc.).

Usage data structure changes:

stats and symbol fields are now omitted when empty
New fields: quantity (float64) and total_usage (float64)
Resources field type changed from map[string]decimal to map[string]interface{} (can contain richer nested data including the Usage struct)

New Endpoint: Instance-Level Billing

Method	Endpoint	Description
`GET`	`/instance-billing/{instance_id}/{start_date}/{end_date}/{interval}`	Billing data for a specific instance

Optional query param: timezone (enum: local, utc).

5. GPUaaS & Pod Management Changes

BREAKING: Pool Worker Connection Details Schema Change

The PoolWorkersConnectionDetails response has changed significantly:

Field	Old Type	New Type	Notes
`pool_subscription_id`	uint (required)	uint (nullable)	Now nullable
`username`	string (required)	string (nullable)	Now nullable
`password`	string (required)	string (nullable)	Now nullable
`port`	int (required)	int (nullable)	Now nullable
`ip`	string (required)	string (nullable)	Now nullable
`region_id`	N/A	uint (required)	New field
`team_id`	N/A	string (required)	New field
`team_name`	N/A	string (nullable)	New field
`pool_id`	N/A	uint (required)	New field
`recipe_id`	N/A	uint (nullable)	New field

Available Resources: New Float-Typed Fields

The GET /gpuaas/pool/{pool_id}/available_resources endpoint is new in this release and uses float64 for CPU, memory, and storage fields. The related MaxDeployableVGpusPerNode response now includes new float32 fields (available_cpu_cores, available_memory_mb, available_storage_mb) alongside the existing integer vgpu_count. If you previously used the internal AllocatableResourcesNode struct (which had integer types for resource fields), note that the new response structs use float types instead.

BREAKING: Subscribed Teams Response Change

GET /gpuaas/pool/{pool_id}/subscribed_teams now:

Includes pool_subscription_id in each entry
The status field now always returns "subscribed" (previously returned the actual pool subscription status)

New Pod Lifecycle Endpoints

Method	Endpoint	Description
`GET`	`/pods`	List all running pods (optional `search` query param)
`POST`	`/pods/provision`	Provision a new pod
`POST`	`/pods/deprovision`	Deprovision a pod
`POST`	`/pods/resize`	Resize a pod (returns recipe execution UUID)
`POST`	`/pods/volume`	Attach/detach a volume to/from a pod
`GET`	`/pods/volume-mount`	Get pod volume mount info
`GET`	`/pods/extract-info`	Extract detailed pod info
`GET`	`/pods/status`	Get pod status
`POST`	`/pods/{pool_subscription_id}/migrate_pods`	Migrate legacy pods

New Pod Management Endpoints (User Panel)

Method	Endpoint	Description
`POST`	`/pods/volume-action`	Mount/unmount volumes on pods
`GET`	`/pods/metrics`	Resource usage metrics for a pod
`POST`	`/pods/{pod_id}/resize/instance-type`	Resize pod instance type
`POST`	`/pods/{pod_id}/resize/ephemeral-storage`	Resize pod ephemeral storage
`GET`	`/pods/{pod_id}/resize/instance-type/options`	Get available resize instance types
`GET`	`/pods/{pod_id}/resize/ephemeral-storage/options`	Get available resize storage options
`GET`	`/pods/{pod_id}/resize/cost-estimate`	Cost estimate for resize

New GPUaaS Resource Endpoints

Method	Endpoint	Description
`GET`	`/gpuaas/pool/{pool_id}/available_resources`	Available CPU, memory, storage per node
`POST`	`/gpuaas/{gpuaas_id}/max_deployable_vgpus/batch`	Batch query max deployable vGPUs across pools
`POST`	`/gpuaas/node/reboot`	Reboot a GPUaaS node

New Service Type: `pod_accelerator`

A new service type pod_accelerator has been added. API clients that parse service type enums must now handle this value in addition to the existing types (cpu_only, cpu_gpu_card, cpu_gpuaas).

GPU Subscription Flow Change

The pool subscription flow now pre-reserves GPUs before provisioning. On recipe failure, GPU assignments are automatically cleaned up. On success, assignments are confirmed. This is transparent to API clients but may affect timing expectations.

GPU Spec Listing Change

GET /gpuaas/cards-specs/all no longer requires the timestamp query parameter. The parameter is ignored if sent. The endpoint now always returns all GPU specs.

6. New Endpoints

Unified Instance Management

Method	Endpoint	Description
`GET`	`/instances/unified`	Paginated listing of all instances (VMs + Pods)
`GET`	`/instances/unified/{instance_id}`	Detailed info for a single unified instance
`GET`	`/instances/unified/{instance_id}/exposed-services`	List exposed services for a pod instance
`POST`	`/instances/unified/{instance_id}/retry`	Retry provisioning a failed VM
`POST`	`/instances/unified/{instance_id}/refresh-connections`	Force-refresh pod connection details

Service Provisioning Flow

Method	Endpoint	Description
`GET`	`/service/i/team-accessible-scenarios`	Scenarios accessible to a team
`GET`	`/service/i/team-accessible-regions`	All regions accessible to a team
`GET`	`/service/i/scenario-compatible-services`	Services under a scenario for a team
`GET`	`/service/i/provisioning-info`	Replaces removed `/service/i/details`
`GET`	`/service/i/storage-blocks`	Compatible storage blocks for a service
`GET`	`/service/i/shared-volumes`	Compatible shared volumes (pod services)
`GET`	`/service/i/additional-disks`	Compatible additional disks for VM services
`GET`	`/service/i/diagnose`	Diagnostic info about service eligibility

Service Options

Method	Endpoint	Description
`GET`	`/service/opts/images`	Image options filtered by service
`GET`	`/service/opts/instance-types`	Instance type options by service
`GET`	`/service/opts/storage-blocks`	Storage block options by service
`GET`	`/service/opts/accelerators`	Accelerators grouped by vendor
`POST`	`/service/opts/accelerator-pools`	Accelerator pool options for a service

Instance IP Retrieval

Method	Endpoint	Description
`POST`	`/instances/ips`	Get IPs for instances updated since a timestamp
`GET`	`/ips/instance/{instance_id}`	Get public IPs for a specific instance

Recipe Execution

Method	Endpoint	Description
`GET`	`/recipes/exec/status/{exec_uuid}`	Get recipe execution status by UUID

7. Removed Endpoints

Method	Endpoint	Replacement
`GET`	`/service/i/details`	Use `/service/i/provisioning-info`
`GET`	`/gpuaas/sub/{sub_id}/pool/{pool_id}/target-url`	Removed (no replacement)
`GET`	`/clusters`	Removed (was already returning 403)
`GET`	`/service/{team_id}/health`	Removed

8. Instance Credentials Response Change

BREAKING

The instance credentials response now includes additional fields and makes all fields nullable:

Old format:

{ "username": "string", "password": "string" }

New format:

{ "ip": "string or null", "username": "string or null", "password": "string or null", "port": 22 }

9. Rate Limiting

A new rate limiting system has been added to all authenticated API endpoints.

Parameter	Value
Default limit	60 requests per minute
Burst allowance	10 requests
Response code when exceeded	HTTP 429 Too Many Requests
Exemption	Full administrator accounts

Action required: Implement retry logic with exponential backoff for HTTP 429 responses in your API clients.

10. New Error Codes

Instance Creation Errors

Error Code	Description
`ERR_VM_OPTIONS_NOT_APPLICABLE_FOR_POD_ACCELERATOR`	VM options provided for a pod_accelerator service
`ERR_POD_OPTIONS_NOT_APPLICABLE_FOR_VM_SERVICES`	Pod options provided for a VM service
`ERR_POD_OPTIONS_MUST_BE_PROVIDED_FOR_POD_ACCELERATOR_SERVICE`	Missing pod_opts for pod_accelerator service
`ERR_VM_OPTIONS_MUST_BE_PROVIDED_FOR_GPU_SERVICES`	Missing vm_opts for GPU VM service
`ERR_INVALID_GPU_CARD_COUNT`	Invalid GPU card count requested
`ERR_EXCEEDED_MAXIMUM_GPU_CARD_COUNT`	GPU count exceeds maximum allowed
`ERR_NO_AVAILABLE_GPU_POOLS_FOR_THE_REQUESTED_SERVICE`	No GPU pools available for the service
`ERR_INSTANCE_CREATION_FAILED`	General instance creation failure
`ERR_NO_COMPATIBLE_IMAGES_FOR_SERVICE`	No images compatible with the selected service

Pod Operation Errors

Error Code	Description
`POD_PROVISIONING_ERR_CODE` (1190004)	Pod provisioning failed
`POD_ATTACH_VOLUME_ERR_CODE` (1190005)	Pod volume attach failed
`POD_DETACH_VOLUME_ERR_CODE` (1190006)	Pod volume detach failed
`POD_DEPROVISIONING_ERR_CODE` (1190007)	Pod deprovisioning failed
`POD_RESIZING_ERR_CODE` (1190008)	Pod resizing failed
`ERR_FAILED_TO_PROVISION_POD`	Failed to provision pod
`ERR_FAILED_TO_DEPROVISION_POD`	Failed to deprovision pod
`ERR_FAILED_TO_EXECUTE_POD_VOLUME_ACTION`	Volume action failed
`ERR_FAILED_TO_EXECUTE_POD_RESIZE`	Resize failed

Billing & Service Errors

Error Code	Description
`ERR_RATE_LIMIT_EXCEEDED`	Rate limit exceeded (HTTP 429)
`ERR_NO_ACCESSIBLE_REGIONS_FOUND_FOR_THE_TEAM`	Team has no accessible regions
`ERR_GPUAAS_NOT_ACTIVE_IN_SELECTED_REGION`	GPUaaS not active in the selected region
`ERR_SERVICE_NOT_GPU_SERVICE`	Service is not a GPU service
`ERR_SERVICE_NOT_POD_ACCELERATOR`	Service is not a pod_accelerator

11. Migration Guide

Step 1: Migrate GPUaaS Pod Creation from Subscription to Instance API

If you create GPUaaS pods via POST /gpuaas/pool/subscribe:

Switch to POST /service/i/create-instance with a pod_accelerator service.
Map your existing fields to the new format (see Section 3 for full field mapping):
- pool_id -> pod_opts.pool_id
- vgpus -> pod_opts.vgpus
- shared_volumes -> pod_opts.shared_volumes
- instance_type_id -> top-level instance_type_id
- ephemeral_storage_block_id -> root_storage_type_id
- image_uuid -> image_hash
- team_id -> top-level team_id
Add new required fields: service_id (a pod_accelerator service), name, workspace_id, region_id.
Remove fields that no longer exist: is_vip_priority, guaranteed_gpu_share_percent, port_guid, ifb_subscription_id.
Replace POST /gpuaas/pool/unsubscribe with instance deletion/deprovisioning through the unified instance endpoints.

Step 2: Update VM Instance Creation Calls

service_id is now required (previously optional) — always provide it.
Replace image_id (integer) with image_hash (string).
Remove cluster_id from requests.
Change workspace_id to UUID format. Change instance_type_id and root_storage_type_id to UUID format (these are now nullable — omit or set null when not applicable).
Move VM-specific fields (additional_disk_*, networks, gpu_card_count) into a vm_opts object. Note: gpu_model is now vm_opts.passthrough_accelerators.
Remove pcidev_ids and recipe_id — these fields no longer exist.

Step 3: Update Billing Response Parsing

Note: The billing endpoint URLs are unchanged. The timezone and region_id parameters were always optional query parameters in practice — the previous Swagger spec was incorrect and has now been corrected. If you generated client code from the old Swagger spec, regenerate it.

Update billing response parsers:
- Remove top-level pci_devices parsing; PCI costs are now within each instance entry under a pci_dev key.
- Handle new per-instance metadata fields (instance_name, nature, model_type, pool_id, pool_name, total_cost).
- Update Resources parsing to handle map[string]interface{} instead of map[string]decimal.
Team/workspace billing now includes pod instances (migrated GPUaaS pods) alongside VMs. Filter by nature field ("vm" or "pod") if you need to separate them. Pod instances have different resource categories (e.g., Subscription Rate, GPU, TFlops) than VMs.
If you use /team-billing/gpuaas-pool/: After upgrade, migrated pool subscriptions are permanently removed from the database, so this endpoint will return empty or reduced results. Switch to the team/workspace billing endpoints and filter by nature: "pod" to retrieve GPUaaS pod costs. Historical billing data for all periods (including pre-upgrade) is available through the team/workspace billing path.

Step 4: Update Billing Estimate Calls

Update POST /calculate-instance-billing-estimate to use the same new UnifiedInstanceCreateOpts request body format.

Step 5: Handle Nullable Fields in Worker Connection Details

If you parse PoolWorkersConnectionDetails, update your code to handle:

username, password, port, ip as nullable (may be null)
New required fields: region_id, team_id, pool_id

Step 6: Update Instance Credentials Parsing

Handle ip, username, password as nullable strings and add handling for the new port field.

Step 7: Replace Removed Endpoints

Replace /service/i/details with /service/i/provisioning-info.
Remove any calls to /gpuaas/sub/{sub_id}/pool/{pool_id}/target-url.

Step 8: Handle Available Resources Response Types

The GET /gpuaas/pool/{pool_id}/available_resources endpoint is new and returns float64 for CPU, memory, and storage fields. The MaxDeployableVGpusPerNode response also now includes new float32 resource fields. Ensure your client models use float types for these fields.

Step 9: Implement Rate Limit Handling

Add retry logic with exponential backoff for HTTP 429 responses. The default limit is 60 requests per minute with a burst of 10.

Step 10: Handle New Service Type

Update service type parsing to handle the new pod_accelerator enum value in addition to cpu_only, cpu_gpu_card, and cpu_gpuaas.

For the complete Swagger/OpenAPI specification, refer to the updated swagger.json and swagger.yaml files included in this release.