Gemini 4 Argon opens to paid API, Pro/Ultra chat, and Ultra Antigravity

Google's Fairwind-only Argon tease is over. Paid surfaces got the model on 13 October 2026. Free chat did not.

Google opens Argon on paid API, Pro/Ultra chat, and Ultra Antigravity

Google DeepMind opened Gemini 4 Argon globally on 13 October 2026 for developers and enterprises on the paid Gemini API (AI Studio, Android Studio, Firebase AI Logic), for Google AI Ultra subscribers in Google Antigravity, and for Google AI Pro and Ultra subscribers in the Gemini app and AI Mode in Search. The API model code is gemini-4-argon.

That follows the 30 September 2026 announcement that limited first access to Fairwind Program cyber defenders. The earlier Frontier piece on that tease is at /articles/gemini-4-argon.

Introductory API pricing is $2 per million input tokens and $10 per million output tokens, with cached input at 95% off. After 30 November 2026 those rates become $4 and $20. Google's footnotes say that schedule applies to the public-access Argon model only.

The DeepMind model card (published October 2026) says Argon is a new frontier model, not a fine-tune of a prior Gemini, and a large-scale sparse mixture-of-experts transformer with native multimodal support for text, vision, and audio inputs and text outputs. Knowledge cutoff is March 2026; the card notes domain lag is possible. Known limits include hallucinations and occasional slowness or timeouts.

Gemini API docs list a unified input-plus-output context window of 1,048,576 tokens. Default synchronous max output is 65,536 tokens; long decode can reach 1,048,576 with Google's continuation protocol. Thinking (low/medium/high), function calling, code execution, search grounding, structured outputs, computer use (preview), and implicit caching are supported. Explicit caching, image generation, audio generation, and Live API are not.

Paid API teams, Ultra Antigravity, and Pro chat users sit closest

Paid Gemini API customers building in AI Studio, Android Studio, or Firebase AI Logic. Google AI Ultra subscribers using Antigravity (desktop, CLI, or Android Studio Canary). Google AI Pro and Ultra subscribers using the Gemini app or AI Mode in Search. Enterprise buyers on Gemini for business paths Google names in the launch posts.

Fairwind applicants seeking Argon without cyber guardrails for authorized defensive work. Up to 10,000 scientists can apply for Antigravity's highest usage tier free for six months, subject to eligibility, capacity, and selection.

Free and lower-tier Gemini chat users are not in this rollout. Ars Technica reports Google declined to say when, or if, Antigravity will open Argon below Ultra. Google is keeping Gemini 3.1 Pro available for middle-tier needs.

A new Gemini 4 MoE with a 1M window and Google's own bench sheet

Versus the Fairwind-only phase: paid API access, Pro/Ultra consumer chat, Ultra Antigravity, and a stated price cliff date (30 November 2026).

Versus Gemini 3 / 3.1 Pro: a new Gemini 4 family member; up to 1M unified tokens with a documented long-decode path; agent surfaces in Antigravity; and a Frontier Safety Framework assessment that cannot rule out biological and cybersecurity Critical Capability Levels or a Stealth/Situational Awareness threshold, with mitigations Google says bring residual risk below those bars. Public API Argon keeps cyber guardrails. Fairwind can still get the unguardrailed cyber SKU under partner terms.

On Google's October 2026 model-card table, Argon leads Vals Index (68.9%), AutomationBench (51.3%), Vals Finance Agent v2 (65.4%), Harvey's Legal Agent Benchmark (19.6%), DeepSWE v1.1 (77.9%), and LVBench (91.7%), and ties GPT-6 Astra on CWE-bench v1 (68.0%). The same table shows rivals ahead on FrontierSWE v2, Terminal-bench 4.0, Terminal-Bench Science 0.1, PostTrainBench, and OSWorld-2.0 (offline partial). Internal Chrome, ADK, xTB, Gemma, and partner anecdotes (Wiz, Isomorphic Labs, Gamow Labs) are Google-attributed.

Free chat, non-Ultra Antigravity, and independent benches stay unsettled

Independent harness replications of Google's benches are not disclosed. Free or Plus chat access is unsettled. Non-Ultra Antigravity access is unsettled. Google has since published the full Safety & Security Report as a 217-page PDF (Developing Gemini 4 Argon responsibly and safely, 13 Oct 2026). FSF summaries remain on the model card. Rest of the Gemini 4 family has no SKUs or dates in these posts. Per-token sticker price is lower than GPT-6 Astra and Claude Fable list rates in Verge coverage, but long-decode and high-thinking traces can erase that advantage. Whether your workload prefers Argon over 3.1 Pro or a rival is not settled by the launch tables.

Trial paid Argon before the 30 November $4/$20 cliff

API teams: trial gemini-4-argon on the paid tier this month, design agents around the 65K default sync ceiling plus the continuation protocol, and model unit economics at $4/$20 before the 30 November cliff. Prefer implicit-cache-friendly prompt shapes; explicit caching is unsupported.

Ultra coding teams: move long-horizon Antigravity loops to Argon; keep 3.1 Pro as failover where terminal and computer-use benches still favor rivals on Google's own sheet.

Pro chat users: use Argon in the Gemini app and Canvas; do not assume Antigravity parity.

Cyber defenders who need unguardrailed defensive capability: apply to Fairwind rather than treating API Argon as the same SKU.