Google Opens Full Access to Gemini 4 Argon

Google has opened full access to Gemini 4 Argon, its newest model from Google DeepMind, the company's in-house AI research lab. Access used to be preview-only. As of October 13, 2026, Argon is available to developers through the API in AI Studio, to enterprise customers, and to Google AI Ultra subscribers in the Gemini app. Argon took first place on the Vals Index, a benchmark that compares models on finance, legal, and coding tasks. It also scored 87% on ERQA, a physical-reasoning benchmark, beating every other top model. Input tokens cost $2 per million, output tokens $10 per million. After November 30, the price rises to $4 and $20.

Google backed up these results with a number from its own engineering practice, not from some abstract test. On the Chrome browser team, Argon helped raise the acceptance rate of agent-written code changes by more than 35% and cut rollbacks by almost a third. That is inside Chrome's 40-million-line codebase. This is a before-and-after comparison within a single internal team, not a head-to-head with other models on the same task. So the Chrome case and the Vals Index ranking actually measure two different things: standing on a cross-model benchmark versus one company's workflow results. At $2 per million tokens, this model suits teams running lots of cheap, repetitive tasks better than it suits those willing to pay top price for the best model on every single query.