Independent newsroom

Thursday, October 1, 2026

The File

Every claim traces to a primary source. Uncertainty is labeled. Corrections are public.

Google DeepMind · Thursday, October 1, 2026

Google announces Gemini 4 Argon

The File desk · Oct 1, 2026, 10:20 AM UTC

Status

Confirmed (Google blog, Koray Kavukcuoglu, Sept. 30, 2026; DeepMind blog index lists the same post)

On Sept. 30, 2026, Google announced Gemini 4 Argon, its next frontier model. The company says it is strong on hard coding work, office work in fields like law and finance, and cyber defense. For now it is rolling out only to trusted cyber defenders through Google’s Fairwind Program while the company works through U.S. government pre-release review and tighter guardrails. Intro prices on the application programming interface are $2 per million input tokens and $10 per million output tokens. Cached input is 95 percent off the input price. After the intro period, prices become $4 and $20. The output token limit rises to 1 million tokens. Google reports DeepSWE v1.1 at 77.9 percent, AutomationBench at 51.3 percent, LVBench at 91.7 percent, and CWE-bench v1 at 68 percent. Broader developer, enterprise, and consumer access is planned after more guardrail work, starting with paid application programming interface customers and Google AI Ultra subscribers.

On Sept. 30, 2026, Google announced Gemini 4 Argon. The post is by Koray Kavukcuoglu.

Google calls it the next frontier model. The company says it is strong on hard coding work, on office work in fields such as law and finance, and on cyber defense.

The first release goes only to trusted cyber defenders, through Google’s Fairwind Program.

Google says that limited rollout runs while the company works through U.S. government pre-release review and tighter guardrails.

Intro prices on the application programming interface are $2 per million input tokens and $10 per million output tokens.

Cached input is 95 percent off the input price.

After the intro period, the prices become $4 per million input tokens and $20 per million output tokens.

The output token limit rises to 1 million tokens.

Google reports these scores: DeepSWE v1.1 at 77.9 percent, AutomationBench at 51.3 percent, LVBench at 91.7 percent, and CWE-bench v1 at 68 percent.

Broader access for developers, companies, and consumers is planned after more guardrail work.

Google says that wider access starts with paid application programming interface customers and Google AI Ultra subscribers.

Google DeepMind’s blog index lists the Gemini 4 Argon post for September 2026.

What is still unknown or disputed

Primary sources

Every claim in this story is drawn from the documents below. If a fetch failed, that is recorded on the card.

  1. Source 1

    Gemini 4 Argon

    Google · September 30, 2026

    Koray Kavukcuoglu, Sept. 30, 2026. Gemini 4 Argon. First release to trusted cyber defenders through the Fairwind Program during U.S. government pre-release review. Intro application programming interface prices $2 per million input tokens and $10 per million output tokens; cached input 95 percent off. Later prices $4 and $20. Output limit 1 million tokens. DeepSWE v1.1 77.9 percent, AutomationBench 51.3 percent, LVBench 91.7 percent, CWE-bench v1 68 percent.

    https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/

  2. Source 2

    DeepMind blog

    Google DeepMind · September 30, 2026

    The DeepMind blog index lists Gemini 4 Argon for September 2026.

    https://deepmind.google/discover/blog