[BOT ISSUE] Google: add missing reasoning_budget flag for gemini-3.5-flash

[github-actions]

Summary

The gemini-3.5-flash entry (Google AI provider) is missing reasoning_budget: true, while its Vertex counterpart publishers/google/models/gemini-3.5-flash already has it. Google's official thinking documentation confirms that gemini-3.5-flash supports adjustable thinking levels (low, medium, high), which maps to reasoning_budget: true in the catalog.

Every other Gemini model that supports thinking levels has reasoning_budget: true in both its Google and Vertex entries:

• gemini-3.1-pro-preview ✅
• gemini-3.1-flash-lite-preview ✅
• gemini-3.1-flash-lite ✅
• gemini-3-flash-preview ✅
• gemini-2.5-pro ✅
• gemini-2.5-flash ✅

Only gemini-3.5-flash is inconsistent — it has reasoning: true but not reasoning_budget: true.

Affected Entry

Model ID	Provider	Current reasoning_budget	Correct Value	Line
gemini-3.5-flash	google	(missing)	true	~6106

The Vertex entry publishers/google/models/gemini-3.5-flash (line ~8002) already has reasoning_budget: true — no change needed there.

All other fields (pricing, format, max_input_tokens, etc.) are correct and unchanged.

Verification

Check	Status	Detail
Cross-source (thinking support)	✅	Google thinking docs: gemini-3.5-flash supports thinking with levels low, medium, high (default: medium) AND Google AI models page: lists gemini-3.5-flash as "most intelligent model" with reasoning capability
Already fixed	✅	Confirmed gemini-3.5-flash entry lacks reasoning_budget in model_list.json
Vertex consistency	✅	publishers/google/models/gemini-3.5-flash already has reasoning_budget: true
Sibling consistency	✅	All other Gemini models with thinking support have reasoning_budget: true in both Google and Vertex entries
Duplicate check	✅	No open issue covers gemini-3.5-flash reasoning_budget

Verification Notes

Field	Source	Notes
Thinking support	Google thinking docs	Lists gemini-3.5-flash with thinking_level support: low, medium, high
Default thinking level	Google thinking docs	Default is "medium" for gemini-3.5-flash
reasoning_budget convention	Catalog convention	Models supporting adjustable thinking/reasoning get reasoning_budget: true

Local Files Inspected

• packages/proxy/schema/model_list.json — gemini-3.5-flash (line ~6106) has reasoning: true but no reasoning_budget; publishers/google/models/gemini-3.5-flash (line ~8002) has both reasoning: true and reasoning_budget: true

{
  "kind": "stale_metadata",
  "provider": "google",
  "models": ["gemini-3.5-flash"],
  "status": "active",
  "model_specs": {
    "gemini-3.5-flash": {
      "format": "google",
      "flavor": "chat",
      "multimodal": true,
      "input_cost_per_mil_tokens": 1.5,
      "output_cost_per_mil_tokens": 9,
      "input_cache_read_cost_per_mil_tokens": 0.15,
      "displayName": "Gemini 3.5 Flash",
      "reasoning": true,
      "reasoning_budget": true,
      "fallback_models": ["publishers/google/models/gemini-3.5-flash"],
      "max_input_tokens": 1048576,
      "max_output_tokens": 65536,
      "available_providers": ["google", "vertex"]
    }
  },
  "source_urls": [
    "https://ai.google.dev/gemini-api/docs/thinking",
    "https://ai.google.dev/gemini-api/docs/models/gemini"
  ]
}

[github-actions]

Prepared catalog updates for gemini-3.5-flash. Updated models: gemini-3.5-flash.

Verification sources: https://ai.google.dev/gemini-api/docs/thinking, https://ai.google.dev/gemini-api/docs/models/gemini
Verified metadata applied:

• gemini-3.5-flash: format=google, flavor=chat, displayName=Gemini 3.5 Flash, max_input_tokens=1048576, max_output_tokens=65536, pricing(input/output per 1M)=1.5/9, input_cache_read_cost_per_mil_tokens=0.15, providers=google|vertex, reasoning=true, reasoning_budget=true, multimodal=true

Included in the daily batch PR: #904