Gemini 4 Argon Now Available: Price, Access and Limits
The short answer
Google made Gemini 4 Argon generally available on October 13, 2026, 13 days after announcing it, at an introductory $2/$10 per million tokens that doubles to $4/$20 on December 1, 2026. Model ID gemini-4-argon. The Fairwind-only restriction that applied from September 30 is gone for paying customers:
| Where | Who gets Argon (Oct 13, 2026) |
|---|---|
| Gemini API (AI Studio, Android Studio) | Paid tier only — not the free tier |
| Gemini Enterprise (“Gemini for business”) | Eligible enterprise subscriptions |
| Google Antigravity | Google AI Ultra subscribers and enterprises |
| Gemini app and AI Mode in Search | Google AI Pro and Ultra subscribers |
| Researchers | Can apply for Antigravity’s highest usage tier free of charge |
API pricing and the November 30 deadline
From Google’s Gemini API pricing page, per million tokens:
| Mode | Input | Output (incl. thinking) | Cached input |
|---|---|---|---|
| Standard, to Nov 30, 2026 | $2.00 | $10.00 | $0.10 |
| Standard, from Dec 1, 2026 | $4.00 | $20.00 | $0.20 |
| Batch / Flex, to Nov 30 → from Dec 1 | $1.00 → $2.00 | $5.00 → $10.00 | $0.05 → $0.10 |
| Priority, to Nov 30 → from Dec 1 | $3.60 → $7.20 | $18.00 → $36.00 | — |
At the introductory rate Argon is priced like the $2 tier (Claude Sonnet 5.5, GPT-6.1 Sol); from December it sits at Claude Opus 5.5’s $4/$20 level. Budget production workloads at $4/$20. Grounding with Google Search includes 5,000 free requests a month across Gemini 3 and newer models, then $14 per 1,000. See current API prices.
Limits that change how you build
- Context: 1,048,576 tokens, unified across input and output.
- Output: 65,536 tokens by default on synchronous calls; up to 1,048,576 with Google’s long-decode mode, meant for very long agentic runs.
- Inputs: text, image, video, audio and PDF; text out. Computer use is in preview.
- Caching: implicit only — explicit cache objects are not supported, so you cannot pin a long prefix the way you can on Gemini 3.x Flash.
How good is it?
Benchmarks disagree. Vals AI ranks Argon first on its Vals Index (68.90% on October 12, 2026; Sonnet 5.5 67.04%, Opus 5.5 66.97%). Artificial Analysis gives it 53, tied with GPT-6 Astra and Claude Fable 5.1, behind Claude Opus 5.5 (58) and Sonnet 5.5 (56). Google’s own numbers show Argon ahead on DeepSWE v1.1 (77.9% vs 74.2% for Opus 5.5) and Zapier’s AutomationBench (51.3% vs 42.5%), and behind Opus 5.5 on Terminal-Bench 4.0 (57.4% vs 66.4%). Google says more than 100,000 of its employees use Argon each month through an internal version of Antigravity.
Who should try it now: teams doing long, professional-knowledge work (legal, finance, large refactors) — the cheap window to November 30 is the time to run your evals. For terminal-heavy coding agents, Claude Opus 5.5 still leads. Head-to-heads: Argon vs Claude Opus 5.5 vs GPT-6 Astra and Argon vs Gemini 3.8 Flash. Background: what Gemini 4 Argon is.
Last verified: October 14, 2026.