Google is reportedly testing a Gemini 4 'Carbon' model that an employee compared to Claude Opus 5.5 at coding
References to Gemini 4 Argon with context options of up to 900K tokens have also appeared in Google's Antigravity tool ahead of a wider release.
According to documents, screenshots and internal chats seen by Business Insider, Google is testing several Gemini 4 variants named Argon, Barium and Carbon. The Decoder reports that Carbon was deployed on Google's internal coding platform, Jetski, in recent days and is said to beat Argon mainly on programming tasks. One employee compared it to Anthropic's Claude Opus 5.5 but said it needs more testing. Another employee said early Argon versions resembled the older Opus 5 on some coding tasks. Google announced Argon on September 30 and gave access first to selected cybersecurity partners. Paid API customers and Google AI Ultra subscribers are expected to follow. Google describes Argon as its model for frontier reasoning, with Flash for speed and volume, Omni for generative media, and Gemma for lightweight open-weight edge workloads. TestingCatalog reports that Antigravity now lists Argon in its model selector with three context settings: 256K tokens by default, 512K at about 1.3 times the quota per turn, and 900K at about 1.8 times. Gemini on the web also now offers low, medium and high thinking-effort settings. None of the reported comparisons have been independently verified.