Google pointed its new model at the drone physics simulator used by Wing, and it found critical bugs that had been sitting in the code since 2018.
That's one of the stories Google is telling about Gemini 4 Argon, which went live on October 13. It's the company's first major frontier release since Gemini 3 in late 2025. The Verge notes that Google spent the months in between trailing rivals from Anthropic and OpenAI.
Cheap, but only if you pay
Google says Argon beats GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5 across coding, cybersecurity, science, finance and law. It also claims the #1 spot on the Vals Index, which covers work like finance, coding, legal and tax.
The API launch price is $2 per million input tokens and $10 per million output tokens, with cached input tokens 95% off. The Verge puts that at around a fifth of what OpenAI's GPT-6 Astra and Anthropic's Fable 5 models cost. Ars Technica reports the rate jumps to $4 and $20 on November 30. The Verge adds that fewer dollars per token doesn't automatically mean a cheaper bill, since models burn different numbers of tokens on the same job.
Access is uneven. Argon is in the Gemini app and AI Mode in Search for Pro ($20 a month) and Ultra (from $99) subscribers. In Antigravity, Google's agentic coding platform, it's Ultra only. Ars says AI Studio is Ultra-only too, while Google's post lists the AI Studio API as open to developers. Google hasn't said if free or Plus users will get it.
What Google says it did with it
Inside Google, Argon in Antigravity helped teams raise agentic code submissions by more than 35% and cut rollbacks by nearly a third. In Chrome, it helped untangle a roughly 6-million-line folder of browser features inside Chromium's 40-million-line codebase. Google says that turned months of work into weeks.
On the Agent Development Kit, Argon worked overnight from a one-paragraph goal and cut in-memory session prep time by 40% in long sessions. Cold imports got 21% faster. Google also got a 2.85x inference speedup on its open Gemma 4 E2B model, from about 37,900 to over 108,000 tokens per second, with no quality drop.
On YouTube comments, Ars reports that removing redundant emoji rendering cut server load by 45%. The Verge says Google described it as halving the computing power. The two accounts don't match exactly.
