[BOT ISSUE] Missing Together AI models: deepseek-ai/DeepSeek-V4-Pro, Qwen/Qwen3.5-397B-A17B

[github-actions]

Gap

Two models available on Together AI's serverless inference platform are missing from packages/proxy/schema/model_list.json:

1. deepseek-ai/DeepSeek-V4-Pro — DeepSeek's flagship V4 Pro model (1.6T total params, 49B active), released April 24 2026. Already available on Together AI serverless with published pricing.
2. Qwen/Qwen3.5-397B-A17B — Qwen's flagship 3.5 model (397B total, 17B active MoE), released February 2026. Available on Together AI serverless with published pricing.

The catalog already tracks the same model families on Together (deepseek-ai/DeepSeek-V3, deepseek-ai/DeepSeek-V3-0324) and both models follow the existing namespace/ModelName ID convention used by Together AI entries.

Verification

deepseek-ai/DeepSeek-V4-Pro

Field	Value	Source
Model ID	deepseek-ai/DeepSeek-V4-Pro	Together AI serverless models
format	openai	Matches existing Together AI DeepSeek entries
flavor	chat	Together AI docs (chat completions)
input_cost_per_mil_tokens	2.10	Together AI serverless models
output_cost_per_mil_tokens	4.40	Together AI serverless models
input_cache_read_cost_per_mil_tokens	0.20	Together AI serverless models
max_input_tokens	512000	Together AI serverless models
available_providers	["together"]	Together AI serverless listing
Cross-source confirmation	DeepSeek V4 Pro confirmed on DeepSeek API pricing and release announcement

• max_output_tokens: not published on Together AI docs — leave unset rather than guessing.
• reasoning: DeepSeek V4 Pro supports thinking mode on DeepSeek's own API; Together AI docs do not confirm this — leave unset.

Qwen/Qwen3.5-397B-A17B

Field	Value	Source
Model ID	Qwen/Qwen3.5-397B-A17B	Together AI serverless models
format	openai	Matches existing Together AI entries
flavor	chat	Together AI docs (chat completions)
input_cost_per_mil_tokens	0.60	Together AI serverless models
output_cost_per_mil_tokens	3.60	Together AI serverless models
input_cache_read_cost_per_mil_tokens	0.06	Together AI serverless models
max_input_tokens	262144	Together AI serverless models
available_providers	["together"]	Together AI serverless listing
Cross-source confirmation	Model confirmed on Hugging Face (released Feb 2026)

• max_output_tokens: not published on Together AI docs — leave unset.
• multimodal: Qwen3.5 is natively multimodal per HuggingFace docs, but Together AI docs do not confirm vision support for this endpoint — leave unset.

Files inspected

• packages/proxy/schema/model_list.json — confirmed neither model ID exists
• git log --oneline -30 -- packages/proxy/schema/model_list.json — no recent commit adds these models

Verification notes

• Together AI serverless models page → confirms model IDs, context windows, and pricing for both models
• DeepSeek API pricing page → independently confirms deepseek-v4-pro exists as a current production model
• Hugging Face model page → independently confirms Qwen/Qwen3.5-397B-A17B exists as a released model
• Fields not published by official sources (max_output_tokens, reasoning, multimodal) are left unset

{
  "kind": "missing_model",
  "provider": "together",
  "models": ["deepseek-ai/DeepSeek-V4-Pro", "Qwen/Qwen3.5-397B-A17B"],
  "status": "active",
  "model_specs": {
    "deepseek-ai/DeepSeek-V4-Pro": {
      "format": "openai",
      "flavor": "chat",
      "available_providers": ["together"],
      "input_cost_per_mil_tokens": 2.10,
      "output_cost_per_mil_tokens": 4.40,
      "input_cache_read_cost_per_mil_tokens": 0.20,
      "max_input_tokens": 512000,
      "displayName": "DeepSeek V4 Pro"
    },
    "Qwen/Qwen3.5-397B-A17B": {
      "format": "openai",
      "flavor": "chat",
      "available_providers": ["together"],
      "input_cost_per_mil_tokens": 0.60,
      "output_cost_per_mil_tokens": 3.60,
      "input_cache_read_cost_per_mil_tokens": 0.06,
      "max_input_tokens": 262144,
      "displayName": "Qwen3.5 397B A17B"
    }
  },
  "source_urls": [
    "https://docs.together.ai/docs/serverless-models",
    "https://api-docs.deepseek.com/quick_start/pricing",
    "https://huggingface.co/Qwen/Qwen3.5-397B-A17B"
  ]
}

[github-actions]

Prepared catalog updates for deepseek-ai/DeepSeek-V4-Pro, Qwen/Qwen3.5-397B-A17B. Added models: deepseek-ai/DeepSeek-V4-Pro, Qwen/Qwen3.5-397B-A17B.

Verification sources: https://docs.together.ai/docs/serverless-models, https://api-docs.deepseek.com/quick_start/pricing, https://huggingface.co/Qwen/Qwen3.5-397B-A17B, https://api-docs.deepseek.com/news/news260424
Verified metadata applied:

• deepseek-ai/DeepSeek-V4-Pro: format=openai, flavor=chat, displayName=DeepSeek V4 Pro, max_input_tokens=512000, max_output_tokens=not provided, pricing(input/output per 1M)=2.1/4.4, input_cache_read_cost_per_mil_tokens=0.2, providers=together
• Qwen/Qwen3.5-397B-A17B: format=openai, flavor=chat, displayName=Qwen3.5 397B A17B, max_input_tokens=262144, max_output_tokens=not provided, pricing(input/output per 1M)=0.6/3.6, input_cache_read_cost_per_mil_tokens=0.06, providers=together

Opened #517 to apply the catalog update.