Back to Technology briefing
Exterior of Google's Googleplex headquarters buildings in Mountain View, California
Technology

Google opens Gemini 4 Argon to paying users and developers

Google on Tuesday made Gemini 4 Argon, its new frontier AI model, available through its API and to paying Gemini subscribers, two weeks after a limited release to cyber defenders. Free users don't get it yet, and the coding tool Antigravity is limited to the top Ultra plan.

By US Brief desk · Updated 2026-10-13T22:13:00-07:00

AI-assisted · US Brief desk · Sources listed below

What happened

Google on Tuesday made Gemini 4 Argon, its newest frontier AI model, generally available, two weeks after it first released the model only to a small group of cyber defenders. Developers can use it through the paid tier of the Gemini API, businesses through Gemini for business and the API, and consumers in the Gemini app and AI Mode in Search on paid plans, Google DeepMind's Koray Kavukcuoglu wrote in a company blog post.

Why it matters

Argon is Google's first major frontier release since Gemini 3 in late 2025, The Verge noted, after months in which OpenAI and Anthropic pulled ahead. Pricing it well below rivals is a direct bid for the developers and businesses building AI agents, and it lands a day after Meta's launch, so three big labs are now competing on price as well as capability.

More context

Who gets it: In the Gemini app, Argon is limited to AI Pro ($20 a month) and AI Ultra (from $99 a month) subscribers, Ars Technica reported. In Antigravity, Google's agent-based coding tool, and in AI Studio, it is available only on Ultra plans. Pro subscribers can still generate code with Argon in the app's Canvas mode, and Google is keeping Gemini 3.1 Pro available for now. The Verge said it is not clear when, or whether, free users will get the model.

What it costs developers: Introductory API pricing is $2 per million input tokens and $10 per million output tokens, with cached input 95% off. After Nov. 30 it rises to $4 and $20. The Verge estimated the introductory rate at about a fifth of the cost of OpenAI's GPT-6 Astra and Anthropic's Fable 5 models, though it noted that models can use different numbers of tokens for the same task. Google says the API allows a combined 1 million input and output tokens.

What Google claims: Google says Argon tops the Vals Index, which measures work in finance, coding, law and tax, and The Verge reported that Google claims it beats GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5 on a range of benchmarks. Internally, Google says Argon raised agentic code submissions in Antigravity by more than 35% while cutting code rollbacks by nearly a third. Ars Technica reported that Argon found redundant emoji rendering in YouTube's comment system that, once removed, cut server load by 45%, and found bugs dating to 2018 in the drone physics simulator used by Alphabet's Wing.

Safety and access: Google says external biology red-teamers could unlock less than 1% of the risky sub-tasks it tests for, and that shortcut attempts on coding tests fell to 2%. It also says an automated monitor reads the agent's reasoning and actions to block unintended workarounds. Google is expanding its Fairwind program, which gives vetted cyber defenders Argon without cyber guardrails, to hundreds more partners. It is also offering 10,000 scientists six months of Antigravity's top usage tier for free, subject to selection.

Timeline: On Sept. 30, Google announced Argon and released it only to Fairwind cyber defenders. On Oct. 4, Bloomberg reported that some people with access doubted its real-world coding, which Google called inaccurate. On Oct. 12, Meta launched its own frontier models, Muse Wonder and Muse Wonder Full. On Oct. 13, Argon went live for paying users and developers.

What's still unknown: Independent benchmark results; whether the coding doubts Bloomberg reported hold up now that more people can test it; when, or whether, free users and lower-tier Antigravity plans get Argon, which Google declined to say, according to Ars Technica; and the rest of the Gemini 4 family, which Google says is still coming.

6 listed sources

References listed by US Brief; a source count is not a verification score.

Editorial sourcing notes

Updated Oct. 13, 2026, ~10:15 a.m. PT for the public launch; the Oct. 4 version covered the limited release and Bloomberg's report of internal doubts. Availability, prices, internal results and safety figures come from Google's Oct. 13 blog posts. Plan prices and the Antigravity limits come from Ars Technica. The comparison with OpenAI and Anthropic models is Google's claim, as The Verge reported, and The Verge's cost comparison is its own estimate; US Brief has not seen independent benchmark results. Ars Technica notes Google is citing the same benchmarks it published in late September.

Like US Brief? You can support it with a tip.