Google released Gemini 4 Argon, its newest frontier AI model, around the world on 13 October 2026 [1]. In its announcement, Google DeepMind said the model is available to developers and businesses through its paid API, and to people who pay for Google’s AI Pro or AI Ultra plans. Free users don’t get it.
Who can use it
Google says Argon is now available through the API in AI Studio and the Gemini platform, in the Gemini app and AI Mode in Search for AI Pro and AI Ultra subscribers, and in Google Antigravity, its coding tool, for AI Ultra subscribers only [1].
Ars Technica reports that AI Pro costs $20 a month and AI Ultra starts at $99 a month [2]. It also reports that in Antigravity and AI Studio, Argon is limited to AI Ultra plans, and that free users are left with Google’s smaller Flash-Lite model. Google told Ars it is keeping Gemini 3.1 Pro around for the time being [2].
What it costs developers
Developers pay per token, the small chunks of text a model reads and writes. Google says Argon launches at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off [1]. That introductory price ends on 30 November 2026. After that, Google says the price rises to $4 per million input tokens and $20 per million output tokens [1]. So anyone building on Argon should plan for costs to double from 1 December.
What Google says the model can do
All of the following figures come from Google and have not been independently checked.
- Benchmarks. Google says Argon ranks first on the Vals Index, which tests models on work in finance, coding, law and tax [1]. On ERQA, a test of reasoning about physical space, Google gives Argon a score of 87% [1].
- Longer tasks. Google describes a “1M unified input-output token limit” through the Gemini API, meaning input and output together can reach about a million tokens [1].
- Coding. Google says Argon in Antigravity helped its teams raise agentic code submissions by more than 35% while cutting code rollbacks by nearly a third [1].
- Speed-ups. Google says Argon helped it make its small open model Gemma 4 E2B run 2.85 times faster on an Nvidia A100 chip using the vLLM software [1].
- Safety. Google says Argon cut the success rate of attacks on the Gray Swan prompt-injection test to 0.7% across 15 adaptive attempts [1]. Prompt injection is when hidden instructions in a web page or file try to hijack an AI model.
Google also says it is expanding its Fairwind Program, which gives vetted cyber defenders access to Argon, to hundreds more partners, and is offering 10,000 scientists the top Antigravity tier free for six months, subject to eligibility, capacity and selection [1].
The cornhole robot
Google says Argon watched a raw video of a person playing cornhole, a game where players throw beanbags at a board with a hole in it, and then wrote simulation code for the movements [1][3]. When that was transferred to Apptronik’s Apollo humanoid robot, Google says the robot sank seven shots in a row in a real-world trial. Google describes this as an internal test, so it was not run or checked by anyone outside the company.
What this means
- Argon is a paid product. You need a paid plan or a paid API account to use it.
- The cheap API price has an end date. The $2 and $10 rates last until 30 November, then they double.
- The headline numbers are Google’s. Outside testers will need time to check the benchmark claims.
Sources
- Google DeepMind (Koray Kavukcuoglu), “Gemini 4 Argon: our next era of frontier intelligence”, 13 October 2026. https://deepmind.google/blog/introducing-gemini-4-argon/
- Ars Technica (Ryan Whitwam), “Gemini 4 Argon begins rolling out in Gemini app and Antigravity, but only if you pay”, 13 October 2026. https://arstechnica.com/ai/2026/10/gemini-4-argon-begins-rolling-out-in-gemini-app-and-antigravity-but-only-if-you-pay/
- Google Developers Blog (Kanishka Rao), “Gemini 4 Argon for Robotics”, 13 October 2026. https://developers.googleblog.com/gemini-4-argon-for-robotics
