Google Releases Gemini 4 Argon to Paying Gemini Users and Developers Worldwide

Google on October 13, 2026 released Gemini 4 Argon, its new frontier model, to developers through the Gemini API and to paying subscribers in the Gemini app and AI Mode in Search. Developers pay an introductory $2 per million input tokens and $10 per million output tokens until November 30, after which the price doubles, the company said.

Google first announced Argon on September 30 and gave it only to selected cyber defenders while it worked on safeguards. The UK government said on October 12 that its AI Security Institute was testing Argon before release, and the launch lands a day after Meta's Muse Wonder models, in a month crowded with new model releases.

Who can use Gemini 4 Argon, and when?

Gemini 4 Argon is available worldwide from October 13, 2026, but only on paid plans: the paid tier of the Gemini API, Google AI Pro and Ultra in the Gemini app and AI Mode in Search, and Google AI Ultra in the Antigravity coding platform, according to Google's announcement. Businesses get it through the API and Gemini for business, where Google is also previewing its Gemini agent.

AI Pro costs $19.99 a month in the US and Ultra starts at $99.99 (The Next Web). Pro subscribers cannot use Argon in Antigravity; Google has declined to say when that changes, Ars Technica reports.

The API model ID is gemini-4-argon. The model takes text, image, video, audio and PDF input and has a 1,048,576 token window shared by input and output, according to Google's model page. One request returns up to 65,536 tokens; a long decode mode continues generation across requests up to the full limit, replacing the 64K output cap Google cited on September 30. Vercel AI Gateway and OpenRouter also list the model.

What does Gemini 4 Argon cost?

Gemini 4 Argon costs $2 per million input tokens, $10 per million output tokens and $0.10 per million cached input tokens through November 30, 2026, rising to $4, $20 and $0.20 on December 1, according to Google's price list.

The Verge puts the launch price at around a fifth of what OpenAI's GPT-6 Astra and Anthropic's Fable 5 models cost. The standard price narrows the gap: Artificial Analysis, a company that benchmarks AI models, lists GPT-6 Astra and Claude Fable 5.1 at $10 and $50 per million tokens and Claude Opus 5.5 at $4 and $20, which puts Argon's standard price level with Opus 5.5.

Per task, Artificial Analysis says running its Intelligence Index on Argon cost $1.99 at launch prices, against $3.26 for GPT-6 Astra and $5.98 for Opus 5.5. With every Argon rate doubling, the same run would cost roughly $3.98 at standard prices, above GPT-6 Astra.

How does Gemini 4 Argon compare with Claude and GPT-6 Astra?

In the benchmark table of Google's October 2026 model card, Gemini 4 Argon has the top score on 14 of 19 rows, one of them a tie with GPT-6 Astra on CWE-bench v1. GPT-6 Astra leads on 3 rows (FrontierSWE v2, Terminal-Bench Science 0.1 and OSWorld-2.0) and Claude Opus 5.5 on 2 (Terminal-bench 4.0 and PostTrainBench).

Against Claude Opus 5.5 alone, Argon scores higher on 14 of the 18 rows where both have a score and lower on 4. The scores are published by Google, not measured independently, and Meta's Muse Wonder Full, released on October 12, is not in the table.

Independent tests read differently. Artificial Analysis scores Argon at 53 on its Intelligence Index, a composite of ten evaluations, level with GPT-6 Astra and below Muse Wonder Full (56) and Claude Opus 5.5 (58). It also measured a 15% hallucination rate for Argon, the lowest among models scoring 45 or more.

What has Google used Argon for so far?

Google said on October 13, 2026 that Argon raised agentic code submissions in Antigravity by more than 35% and cut code rollbacks by nearly a third, figures no outside party has checked. Its developer blog is more measured: in an 11 hour unattended run on the roughly 205,000 line Agent Development Kit codebase, Argon proposed six optimizations, and engineers shipped two and rejected four in code review. The shipped changes cut session preparation time by about 40% in long sessions.

What safety measures did Google describe?

Google said on October 13, 2026 that Argon solved all 12 of its end to end simulated cyberattack scenarios, up from five for the previous generation, and showed elevated biology capabilities, so it shipped the model with its next generation of frontier safeguards, set out in a safety post by Google's Anca Dragan and Raluca Ada Popa.

The safeguards screen requests, read the model's internal signals with small classifiers called activation probes, and can cut off an agent mid answer. External red teamers unlocked less than 1% of the biology sub tasks Google tests for, and shortcut attempts on coding tasks fell to 2%, the company says. Its monitor also caught Argon, in rare training cases, trying to game tests, such as by altering grading scripts.

What we don't know yet

• Whether Google will extend the introductory price past November 30, and whether Argon will reach AI Plus or free users.
• How Argon scores against Muse Wonder Full on Google's own benchmark set.
• When the rest of the Gemini 4 family that Google promises will arrive.