Google’s Gemini 4 “Carbon” model is matching Anthropic’s Opus 5.5 on coding benchmarks — and it hasn’t launched yet. Internal testing shows Carbon, one of several variants under the Gemini 4 umbrella, outperforming the still-unreleased “Argon” model specifically on programming tasks. It’s running now on Google’s internal Jetski coding platform.
No public release date exists. No pricing. No availability window. Just code, comparisons, and quiet urgency inside Mountain View. <br><br>Carbon isn’t a standalone product.
It’s a checkpoint, likely an update within the Argon family, not a new tier. Internal docs refer to the current Argon as “Barium-B,” and one employee reportedly called Carbon “Gemini pro next model,” suggesting it’ll ship under the Argon branding. That fits Google’s stated model taxonomy: Argon for frontier reasoning, Flash for speed and volume, Omni for generative media, Gemma for lightweight edge workloads. <br><br>So what changes for users?
Nothing yet. But the signals are sharp. The Gemini app already shows a new “Automatic” mode for existing 3-series models and lets users dial reasoning intensity from low to high. Google AI Studio has surfaced an “Ultra” mode promising “advanced skills and tools.” Logan Kilpatrick, Google’s head of developer relations for AI, confirmed the team is actively optimizing for Argon, meaning Carbon’s gains aren’t just lab noise. <br><br>The timing matters.
Anthropic’s Opus 5.5 landed earlier this year as its strongest coding model. Google matching it before Argon ships suggests recursive self-improvement isn’t just a slogan. It’s shipping velocity. Vedant Misra of Google DeepMind alluded to that on X, replying to the Business Insider report with: “Have you heard of recursive self improvement.” <br><br>This isn’t a launch.
It’s a pivot point. Argon was expected to be Google’s answer to Opus and OpenAI’s Astra. Carbon implies that answer may arrive faster, and sharper. Than anyone anticipated.
Next step: official confirmation of Argon’s release date. Or, quietly, Carbon slipping in as its first update.
ADFiled by The AI Desk
Models, assistants and the companies and chips behind them, reported from what was released and what was claimed, with the difference kept clear.
More from this desk →
Be the first to comment
Join the argument. No password, just your email or a passkey.