Gemini 4 Carbon Reportedly Matches Opus 5.5 on Coding

Gemini 4 Carbon model facing off against the Opus system in a coding duel

Reports indicate Google is testing a Gemini 4 variant called Carbon that performs on par with Anthropic’s most capable coding model, Opus 5.5, giving developers a preview of where the next Gemini generation may land on programming workloads.

What internal tests show about Carbon

Documents, screenshots, and internal chats reviewed by Business Insider describe several Gemini 4 variants in development under the names Argon, Barium, and Carbon. Carbon was deployed on Jetski, Google’s internal coding platform, over the past few days and is said to outperform Argon mainly on programming tasks. One employee compared Carbon to Anthropic’s strongest coding model, Opus 5.5, though the model still requires more testing before any public claims are made. Earlier Argon builds, by contrast, reminded another employee of the older Opus 5 on some coding tasks, even though internal feedback on Argon was positive overall.

How Carbon fits inside the Gemini 4 lineup

Early speculation cast Argon as a Flash model tuned for speed over peak performance. Google’s announcement of its Gemini agent for Google Workspace has put that idea to rest. The company now maps its models to specific roles: Argon for frontier reasoning, Flash for speed and volume, Omni for generative media, and Gemma for lightweight, open-weight edge workloads. Under that framing, Argon is Google’s most capable reasoning model, comparable to Anthropic’s Opus line or OpenAI’s Astra. Carbon and Barium appear to be checkpoints or updates within the Argon family rather than separate tiers. One employee internally labeled Carbon as the next Gemini pro model, suggesting it would ship under the Argon name. Internal documents show the Argon build that has now been unveiled was previously called Barium-B. Whether Carbon ships as an Argon update or a standalone model remains unclear.

Why the iteration cycle is speeding up

The pace of updates mirrors what Google has already done with the Flash models and points to a broader shift in how frontier labs are building their next systems. Google DeepMind staff have pointed to recursive self-improvement, the practice of using AI to build better AI, as a driver behind the rapid turnaround. OpenAI and Anthropic have reported similar progress in their own model pipelines.

What this means for coding workflows

For developers who already use Gemini for code generation, the practical question is how soon a Gemini 4 tier tuned for programming will be available through the Gemini API, AI Studio, or the Gemini app. Internal signals suggest that work is already underway on the application layer. Google has been retooling its apps ahead of any official Gemini 4 launch. The Gemini app now shows a new Automatic mode for the 3-series models and adjustable reasoning intensity from low to high. Google AI Studio has surfaced a new Ultra mode promising advanced skills and tools. The team has confirmed it is working to get the most out of Argon, which would include any Carbon-derived improvements once they ship.

What to watch next

Three developments will determine whether the Carbon reports translate into public capability gains. First, the official Gemini 4 launch date and naming, since the spec was never confirmed as a public codename. Second, third-party benchmarks on coding tasks such as HumanEval, SWE-bench, and LiveCodeBench that can verify or contradict the internal Opus 5.5 comparison. Third, pricing and rate limits through the Gemini API, which determine whether the new model is usable at production scale. Until then, Carbon remains a known-internal signal that Google is closing the coding gap with Anthropic’s top tier, even if it has not yet shipped to external developers.

FAQ

What is Google’s Gemini 4 ‘Carbon’?

Carbon is the reported codename for a Gemini 4 variant Google has been testing on its internal coding platform Jetski. Internal feedback suggests it outperforms Argon on programming tasks and is comparable to Anthropic’s Opus 5.5, though it still needs more testing.

How does Carbon relate to Argon and Barium?

Carbon and Barium appear to be checkpoints or updates within the Argon family rather than separate model tiers. Internal documents indicate that the now-unveiled Argon was previously called Barium-B, and one employee referred to Carbon as the next Gemini pro model that would ship under the Argon name.

Why is Google iterating on Gemini models so quickly?

The rapid update pace is similar to what Google has done with the Flash models and reflects progress in recursive self-improvement, the practice of using AI to build better AI. OpenAI and Anthropic have reported similar advances in their own model pipelines.

Related coverage

This article summarizes reporting from the-decoder.com. See our editorial disclaimer for how our articles are produced.

Google’s Gemini 4 ‘Carbon’ Model Reportedly Matches Anthropic Opus 5.5 on Coding Tasks