On October 13, 2026, Google made Gemini 4 Argon broadly available, two weeks after announcing it on September 30 with access limited to vetted defenders in its Fairwind Program. Developers and enterprises can use it through the Gemini API paid tier and AI Studio, and Google AI Pro and Ultra subscribers get it in the Gemini app and AI Mode in Search. Google also expanded the Fairwind Program to hundreds more cyber-defense partners, who can get access without the cyber guardrails.
Introductory pricing is $2 per million input tokens and $10 per million output tokens, with cached input 95 percent off, rising to $4 and $20 after November 30, 2026. The API offers what Google calls an industry-leading 1-million-token unified input-output limit. Google reports Argon ranked first on the Vals Index, scored 87 percent on the ERQA embodied-reasoning benchmark, and showed 0.7 percent attack success on Gray Swan’s indirect prompt injection test.
The same day Google published “Gemini 4 Argon for Robotics,” presenting the model as a general physical-reasoning model that can control robots, build simulation twins from camera feeds, video or sketches, and improve real-robot task success from 56 to 75 percent after 4,500 simulated episodes. The Gemini app update also brought Antigravity’s agentic coding interface to the Gemini macOS app and connections to more than 40 apps.
Why it matters: Argon is the first frontier model released in two stages, defenders first and the public second, a pattern that turns cyber capability into a gate on deployment rather than an afterthought.
What it does not show: the benchmark, safety and robotics figures are Google’s own, and the two-week gap between stages is not evidence of what the defender-only period found or fixed.