Google’s new Gemini 4 Carbon took first place on the leaderboard run by Artificial Analysis, an independent AI benchmarking group. It scored 59, one point ahead of Claude Opus 5.5 and two points ahead of OpenAI’s GPT-6 Astra.
The release came two weeks after Argon, Google’s previous top model. Google CEO Sundar Pichai reiterated that the company plans to release models periodically.
For now, access is restricted to a small group of cybersecurity defenders, with Google also participating in voluntary testing with the US government before a wider release.
- It scored 88.2% on SEC-bench Pro, a benchmark for finding security vulnerabilities in real software, with Google’s safeguards switched off.
- Google’s Chrome team reported that its security fixes performed more than 10 times better than those from another lab’s leading model.
- In a simulated vending machine business, it finished with $16,484 and declared its priority was to obliterate the competition.
The lead is narrow: one point on a benchmark where 60% of tests remain private to reduce the risk of models being trained directly on them. Claude Code still leads the separate coding-agent ranking, scoring 62 against 59.
On the same day, Alphabet’s stock fell about 10% after the company missed both revenue and profit expectations for the first time in nearly four years.
