Gemini 4 Argon: Who Gets It, When, and What It Costs

Short answer

You can't use Gemini 4 Argon because Google hasn't released it to the public. It's going to vetted cyber defenders first.Google announced Argon on September 30, 2026. It is rolling out first to trusted cyber defenders in its Fairwind Program. Paid API customers and Google AI Ultra subscribers come next, under a section Google titles "Rolling out soon". Google gives no date for that step. As of today, the Gemini API models page doesn't list any Gemini 4 or Argon model. The newest models there are from the 3.8 generation.

Source: Gemini 4 Argon: our next era of frontier intelligence (Google) · 2026-10-12; Gemini models (Gemini API docs) · 2026-10-12

What Google actually announced

Google calls Argon a frontier model for software engineering, enterprise work like legal and finance, and cyber defense. The headline spec is output length. Argon can return up to 1M tokens, up from the previous 64K limit. Google also says Argon can "autonomously find, validate, and patch critical software vulnerabilities." Trusted defenders and Google's internal teams will get a version without cyber guardrails.

Source: Gemini 4 Argon: our next era of frontier intelligence (Google) · 2026-10-12

A baffled person holds a paper medal beside an empty pedestal where a product should be. — Google announced Argon with strong benchmark results, but there is no product available to use yet.

Does it really beat OpenAI and Anthropic?

On some tests it does, by small margins, and many of the numbers come from Google. 9to5Google reports Argon at 77.9% on DeepSWE v1.1, ahead of Claude Opus 5.5 at 74.2% and GPT-6 Astra at 74.1%. On CWE-bench v1, Google DeepMind's own chart shows Argon tied at 68% with GPT-6 Astra and Grok 4.7, with Opus 5.5 at 67%. Engadget cites Artificial Analysis, which has Argon matching GPT-6 Astra's Intelligence Index score at lower cost per task, with the lowest hallucination rate among leading models at 15%. Help Net Security points out that the vulnerability-discovery results come from Google's internal tests.

Source: Google announces Gemini 4 Argon as its new frontier model (9to5Google) · 2026-10-12; Gemini 4 Argon with cybersecurity defense capabilities (Google DeepMind) · 2026-10-12; Google Gemini 4 model Argon (Engadget) · 2026-10-12; Google Gemini 4 Argon (Help Net Security) · 2026-10-12

Our take: a three-point lead you can't check

Beating Opus 5.5 by about 3.7 points on a coding benchmark is a real result. It is not a knockout, and a tie on CWE-bench is still a tie. The bigger problem is that nobody outside Fairwind can test these claims on their own work. Until you can call the model, the benchmark chart is marketing. Our view is that Opus 5.5 and GPT-6 Astra users have no reason to switch today, because you can't switch to a model that isn't for sale.

Who gets it, in order

A panicking person waits at the back of a rope line while a guard admits only the first person. — Google plans to give Argon to vetted cyber defenders first, so most people will be waiting for broader access.

1. Fairwind Program partners (now)Google DeepMind gives priority to governments and critical-infrastructure operators such as healthcare, telecom, energy and finance. Core technology platforms qualify too, and academic labs doing defensive benchmarking can apply. Every applicant is vetted. Inside a partner organization, only internal cybersecurity, incident response or penetration testing teams can use Argon.
2. Paid API customers and Google AI Ultra subscribers ("soon")Google names these two groups as the start of the broader release. It gives no date and doesn't say which AI Ultra plan gets access or whether Argon appears in the Gemini app.
3. Developers, enterprises and consumers generallyGoogle says it will release Argon "as soon as possible" in phases. Help Net Security reports that Google is still adjusting guardrails based on feedback from early testers.

Source: Fairwind Program (Google DeepMind) · 2026-10-12; Gemini 4 Argon: our next era of frontier intelligence (Google) · 2026-10-12; Google Gemini 4 Argon (Help Net Security) · 2026-10-12

we do not allow partner organizations to share, redistribute, or sell access to our frontier models.

Google DeepMind, Fairwind Program page (no date shown; read 2026-10-12)

Source: Fairwind Program (Google DeepMind) · 2026-10-12

Why this matters to you: Fairwind members can't resell or share Argon access. If anyone offers you early access to Argon right now, it isn't a Fairwind channel Google allows. Partners also have to track which employees use the model.

Source: Fairwind Program (Google DeepMind) · 2026-10-12

What the API will cost

Google lists introductory API rates of $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper. A footnote says the rates rise to $4 and $20 once the introductory period ends. Google doesn't say how long that period lasts. Google hasn't announced a consumer price for Argon either, because AI Ultra access is the planned route.

Source: Gemini 4 Argon: our next era of frontier intelligence (Google) · 2026-10-12

Our take: watch the 1M-token output pricing

A 1M-token output limit priced per output token makes long jobs expensive. At the standard rate, a single response that uses the full limit would cost about $20 in output alone. The introductory price makes the launch look cheap, but it ends on a date Google hasn't announced. Budget for the $4/$20 rates from the start.

What to do this week

Developers: check the Gemini API models page for a Gemini 4 or Argon model ID before you plan any migration. It isn't listed yet.
Security teams at eligible organizations: apply through the Fairwind interest form on Google DeepMind's site.
AI Ultra subscribers: Google hasn't confirmed which Ultra plan gets Argon or when, so don't upgrade only for Argon.
Anyone offering Argon access for sale: Fairwind terms forbid resale, so skip it.

Source: Gemini models (Gemini API docs) · 2026-10-12; Fairwind Program (Google DeepMind) · 2026-10-12; Gemini 4 Argon: our next era of frontier intelligence (Google) · 2026-10-12

Our verdict

Argon's best feature is overrated in the hype, and its rollout is the real story. Giving defenders the model without guardrails before attackers can get a comparable one is a defensible choice. Calling that a consumer launch isn't. For now, Argon is for enterprises and governments, and subscribers have to wait. The 1M-token output and the patching claims are worth getting excited about once people outside Google can test them.

Optional: see third-party seller listings for Gemini plans. These are seller listings, not official Google access and not an endorsement. Buying one won't get you Argon.. That link carries our partner code Sk52g, which is how a purchase is attributed to this site. Promotion code UPD3S.

Sources (all read 2026-10-12)

Gemini · Compare price & terms →

Gemini 4 Argon Is a Scoreboard, Not a Product (Yet)