[vertex-ai][google] Add Gemini 4 Argon, fix Argon region/tier pricing

[naresh4dev]

What

Adds Gemini 4 Argon (gemini-4-argon, GA 2026-10-13) to the Gemini Developer API provider, and completes/corrects the Vertex AI entry that landed in 05ce0d7.

	general/	pricing/
vertex-ai	fixed params	completed region × mode matrix
google	added	added

Why

Two bugs in the existing Vertex entry:

1. Non-global regions and priority/flex/batch were all billed at the global standard rate. The entry had a single flat pay_as_you_go block with no custom_pricing.regions, unlike every other Gemini model in the file. Priority traffic was undercharged 44%, batch/flex overcharged 100%.
2. gemini-4-argon-cyber-permissive was given the promotional rate. The pricing page footnote says "Promotional pricing does not apply to Gemini 4 Argon Cyber Permissive" — it was undercharged by 50%.

Params were also wrong for this model family: Argon returns HTTP 400 if thinking_budget is set and only accepts thinking levels, but the entry exposed thinking.budget_tokens.

Pricing

Promotional rate (in effect through 2026-11-30), $/1M ÷ 10000 → cents/token:

Mode	Global (in / cached / out)	Non-global
standard	$2 / $0.10 / $10	$2.2 / $0.11 / $11
batch, flex	$1 / $0.05 / $5	$1.1 / $0.055 / $5.5
priority	$3.6 / $0.18 / $18	$3.96 / $0.198 / $19.8

Cyber Permissive takes the list rate (exactly 2× the above). Grounding is $14/1k queries → 1.4.

Note

On 2026-12-01 the promo lapses and all Argon rates double. A follow-up is needed then; this PR deliberately matches the existing gemini-3.8-flash convention of carrying the currently-effective rate.

Config notes

• max_tokens: 65536 — the per-request synchronous cap. The docs' "Maximum output tokens: 1,048,576" is only reachable via long-decode continuation_token across multiple requests, so 65536 is the right per-call ceiling.
• reasoning_effort (low/medium/high, default medium) replaces thinking.budget_tokens. The gateway already maps this to thinkingConfig.thinkingLevel for non-2.5 Gemini models, so no gateway change is needed.
• removeParams: temperature, top_p (+top_k on google) — Argon ignores custom sampling values.
• cache_read_input_token only, no cache_write — implicit caching only (6,144-token min); explicit caching is unsupported.
• google keys are -lte-128k/-gt-128k because that provider's handler unconditionally appends a context-tier suffix. Argon isn't context-tiered, so both tiers are identical — same as gemini-3.8-flash.

Testing

• jq empty on all 4 files
• python3 scripts/check_duplicate_keys.py → no duplicates
• npm run format / npx prettier --check → clean
• npm run lint → 2653 errors, all pre-existing in general/scenario-ai.json; identical count on clean origin/main, none in these 4 files
• Script-verified all 48 price values round-trip against the official $/1M tables

Sources

• https://cloud.google.com/vertex-ai/generative-ai/pricing
• https://ai.google.dev/gemini-api/docs/pricing
• https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/4-argon
• https://ai.google.dev/gemini-api/docs/models/gemini-4-argon

[Copilot]

🟢 Approval recommended

The configurations consistently implement the documented parameters and pricing calculations.

0 open findings

What changed in this PR

Adds Gemini 4 Argon support for Google and corrects Vertex AI configuration and pricing.

Changes:

• Adds Google model metadata and pricing.
• Corrects Argon parameters and Cyber Permissive rates.
• Adds Vertex AI regional and execution-mode pricing matrices.

File	Description
general/​google.json	Adds Gemini 4 Argon capabilities and parameters.
general/​vertex-ai.json	Corrects supported Argon parameters.
pricing/​google.json	Adds standard and batch pricing tiers.
pricing/​vertex-ai.json	Corrects rates and adds regional mode pricing.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.