[BOT ISSUE] Vertex: add missing DeepSeek V3.2, V3.1, and R1-0528 MaaS entries

[github-actions]

Summary

Google Cloud Vertex AI offers DeepSeek models as managed MaaS (Model-as-a-Service) open model APIs, but the catalog has no publishers/deepseek-ai/models/ entries. DeepSeek models exist in the catalog under Bedrock (deepseek.v3.2, deepseek.r1-v1:0) and Together/Baseten (deepseek-ai/DeepSeek-V3, deepseek-ai/DeepSeek-R1) naming conventions, but Vertex MaaS entries are entirely absent.

Missing Models

Model ID	Display Name	Context	Max Output	Stage
publishers/deepseek-ai/models/deepseek-v3.2-maas	DeepSeek V3.2	163,840	65,536	GA
publishers/deepseek-ai/models/deepseek-v3.1-maas	DeepSeek V3.1	—	—	Listed
publishers/deepseek-ai/models/deepseek-r1-0528-maas	DeepSeek R1 (0528)	—	—	Listed

V3.2 specs are confirmed from its model card page. V3.1 and R1-0528 model IDs follow the same -maas naming convention and are listed on the official MaaS overview but their detail pages could not be scraped for exact specs.

Suggested Catalog Entry (V3.2)

"publishers/deepseek-ai/models/deepseek-v3.2-maas": {
  "format": "openai",
  "flavor": "chat",
  "displayName": "DeepSeek V3.2",
  "max_input_tokens": 163840,
  "max_output_tokens": 65536,
  "available_providers": [
    "vertex"
  ]
}

Notes:

• Format is "openai" — Vertex MaaS models use OpenAI-compatible request payloads (API model string: deepseek-ai/deepseek-v3.2-maas)
• Pricing omitted — the Vertex pricing page lists generic deepseek-v3 at $0.50/$1.50 per 1M tokens, but this may not map exactly to deepseek-v3.2-maas; verify from individual model pricing before setting
• DeepSeek R1-0528 should include "reasoning": true when added

Sources

• Vertex AI Partner/Open Models listing — lists DeepSeek-V3.2, DeepSeek-V3.1, DeepSeek-R1-0528, and DeepSeek-OCR under Open Models > DeepSeek
• DeepSeek-V3.2 model card (Vertex AI) — confirms model ID deepseek-v3.2-maas, 163,840 context, 65,536 max output, GA stage
• Vertex AI Console: DeepSeek-V3.2 — confirms publisher path publishers/deepseek-ai
• Vertex AI Pricing — lists DeepSeek models with per-token pricing

Verification Checklist

Check	Status	Detail
Cross-source	✅	V3.2 confirmed on partner listing page, model card page (exact ID + specs), AND console URL (publisher path); V3.1 and R1-0528 confirmed on partner listing page
Already fixed	✅	grep for publishers/deepseek in model_list.json returns zero matches
ID format	✅	Follows -maas suffix convention from existing publishers/meta/models/llama-3.3-70b-instruct-maas
Duplicate check	✅	No open issue covers DeepSeek Vertex MaaS entries

Verification Notes

Field	Source	Notes
Model ID deepseek-v3.2-maas	Vertex model card + console URL	GA stage, API format: deepseek-ai/deepseek-v3.2-maas
Context 163,840 / output 65,536	Vertex model card	Confirmed for V3.2 only
format: "openai"	Vertex MaaS docs	"MaaS models expect OpenAI-compatible request payloads"
V3.1 and R1-0528 IDs	Inferred from naming convention	Same -maas suffix pattern; listed on partner models page but detail pages not scraped
DeepSeek-OCR	Excluded	OCR-only model, not a chat/completion model

Local Files Inspected

• packages/proxy/schema/model_list.json — grep for publishers/deepseek returns 0 matches; existing DeepSeek entries exist under Bedrock (deepseek.v3.2 at line ~9186) and Together/Baseten (deepseek-ai/DeepSeek-V3 at line ~9931) naming conventions

{
  "kind": "missing_model",
  "provider": "vertex",
  "models": ["publishers/deepseek-ai/models/deepseek-v3.2-maas", "publishers/deepseek-ai/models/deepseek-v3.1-maas", "publishers/deepseek-ai/models/deepseek-r1-0528-maas"],
  "status": "active",
  "model_specs": {
    "publishers/deepseek-ai/models/deepseek-v3.2-maas": {
      "format": "openai",
      "flavor": "chat",
      "displayName": "DeepSeek V3.2",
      "max_input_tokens": 163840,
      "max_output_tokens": 65536,
      "available_providers": ["vertex"]
    },
    "publishers/deepseek-ai/models/deepseek-v3.1-maas": {
      "format": "openai",
      "flavor": "chat",
      "displayName": "DeepSeek V3.1",
      "available_providers": ["vertex"]
    },
    "publishers/deepseek-ai/models/deepseek-r1-0528-maas": {
      "format": "openai",
      "flavor": "chat",
      "displayName": "DeepSeek R1 (0528)",
      "reasoning": true,
      "available_providers": ["vertex"]
    }
  },
  "source_urls": [
    "https://docs.cloud.google.com/vertex-ai/generative-ai/docs/partner-models/use-partner-models",
    "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/maas/deepseek/deepseek-v32",
    "https://cloud.google.com/vertex-ai/generative-ai/pricing"
  ]
}

Summary: Three DeepSeek MaaS models are available on Vertex AI but have no catalog entries. V3.2 has fully confirmed specs (GA, 163K context, 65K output); V3.1 and R1-0528 are listed but need detail page verification for exact specs.

[github-actions]

Autofix could not safely resolve this ticket.

Autofix skipped because the issue does not yet contain enough verified metadata to safely apply catalog changes for publishers/deepseek-ai/models/deepseek-v3.2-maas, publishers/deepseek-ai/models/deepseek-v3.1-maas, publishers/deepseek-ai/models/deepseek-r1-0528-maas. Refusing to update publishers/deepseek-ai/models/deepseek-v3.2-maas without explicit location metadata

Verified information from this run:

• Provider: vertex
• Parsed models: publishers/deepseek-ai/models/deepseek-v3.2-maas, publishers/deepseek-ai/models/deepseek-v3.1-maas, publishers/deepseek-ai/models/deepseek-r1-0528-maas
• Official source URLs already on the ticket: https://docs.cloud.google.com/vertex-ai/generative-ai/docs/partner-models/use-partner-models, https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/maas/deepseek/deepseek-v32, https://console.cloud.google.com/vertex-ai/publishers/deepseek-ai/model-garden/deepseek-v3.2-maas, https://cloud.google.com/vertex-ai/generative-ai/pricing, https://huggingface.co/deepseek-ai/DeepSeek-V3.1, https://cloud.google.com/blog/products/ai-machine-learning/deepseek-r1-is-available-for-everyone-in-vertex-ai-model-garden

To make autofix work on a future run, update the machine-readable metadata block for this missing model issue and include the verified fields needed for the catalog change, such as:

• model_spec for single-model issues or model_specs for multi-model issues
• displayName
• parent for non-base dated snapshots or location-scoped variants when official sources verify the relationship
• pricing fields
• max_input_tokens
• max_output_tokens
• available_providers
• deprecated / deprecation_date
• locations when the provider/model requires explicit location metadata