Google Staff Test New Gemini 4 'Carbon' Build on Jetski

Google employees are testing a new Gemini 4 variant internally codenamed Carbon, deployed across the company's Jetski development environment in recent days, Business Insider reports.

One Google developer told the outlet that Carbon's coding performance "feels comparable to Anthropic's Opus 5.5." That is one anonymous staffer, speaking to one reporter, about an internal build nobody outside the company has evaluated. It is the kind of quote that drives procurement conversations anyway.

The naming matters because Google's internal codenames do not match its public ones. Business Insider reports that the Barium-B build was the one ultimately selected for public launch as Gemini 4 Argon. Carbon is the next step along that line. Its final public designation is unconfirmed, and Google declined to comment on specific internal roadmaps.

Carbon is landing while Argon itself is still being argued over inside Google. Benchmark scores look strong; some employees say the model is less convincing when they actually use it for coding work. Our tracker has logged 204 Google AI stories over the last 90 days, and this week's run alone — a Workspace MCP v2 release and an AI Edge Foresight launch — reads like a company moving faster than its public roadmap.

No public benchmark for Carbon exists. No release date. No second outlet has confirmed the Opus 5.5 comparison.