Google's Gemini 4 Carbon reportedly matches Opus 5.5 on coding

Google hasn't launched Gemini 4 Argon yet, but internal documents suggest a faster follow-up is already in testing — and early impressions put it level with Anthropic's best coding model.

Google is testing a Gemini 4 variant called "Carbon" that employees say performs on par with Anthropic's Opus 5.5 on programming tasks. Business Insider, citing internal documents, screenshots, and chats, reports that Google deployed Carbon on Jetski — its internal coding platform — over the past few days, with at least one employee comparing its coding ability to Opus 5.5, Anthropic's strongest model. Carbon is one of several internal Gemini 4 variants alongside Argon (the announced frontier model) and Barium, whose Barium-B checkpoint was selected as the public Argon release. It likely ships as an update within the Argon family rather than a separate tier; one employee called it internally the "Gemini pro next model."

The timing is what makes this more than a leak. Argon only just arrived, and Carbon's arrival in internal testing within days suggests Google's model iteration loop is accelerating. DeepMind employee Vedant Misra responded to the report on X with "Have you heard of recursive self improvement" — a nod to the idea that AI is increasingly building better AI. OpenAI and Anthropic have both reported similar dynamics in their own pipelines, and Google recently mapped its lineup explicitly: Argon for frontier reasoning, Flash for speed, Omni for media, Gemma for edge. A Carbon drop would slot straight into that frontier slot before Argon even reaches general availability. We covered Argon's coding reputation before — Google employees question Gemini 4 Argon's real-world coding — and Carbon reads like the direct answer to those internal doubts.

Worth keeping the report proportionate: the Opus 5.5 comparison comes from one employee and, by Business Insider's own account, still needs more testing. Google has not commented. What to watch: whether Carbon ships as an Argon update or stands alone, and whether any of it shows up on public benchmarks before the Gemini 4 launch date lands.

Is AI building the next generation of AI models faster than labs can name them — or is one employee's chat message being over-read? Tell us in the comments.