Google is reportedly testing a new variant of its upcoming Gemini 4 super intelligence (SI) model that matches the coding performance of Anthropic’s Opus 5.5, according to internal documents and employee chats seen by Business Insider. The variant, codenamed Carbon, has been deployed on Google’s internal coding platform, Jetski, where it is said to outperform the frontier reasoning model Argon on programming tasks.
What Happened
Documents and screenshots indicate that Google is evaluating several Gemini 4 variants, including Argon, Barium, and Carbon. While Argon serves as Google’s frontier reasoning SI model, Carbon appears to be a specialized update or checkpoint within the Argon family. One employee compared Carbon directly to Anthropic’s strongest coding SI model, Opus 5.5, noting that while it performs comparably, it still requires further testing. In contrast, early versions of Argon reminded another employee of the older Opus 5 model on certain coding tasks, although internal feedback on Argon was generally positive.
Google has clarified its model mapping strategy in the announcement of its Gemini SI agent for Google Workspace. The company states: "Argon for frontier reasoning, Flash for speed and volume, Omni for generative media, and Gemma for lightweight, open-weights edge workloads." This confirms Argon as the most capable reasoning SI model, comparable to Anthropic’s Opus or OpenAI’s Astra. Internal documents suggest Carbon may ship as an update under the Argon name, with one employee referring to it as the "Gemini pro next model," though whether it will be a standalone release remains unclear.
Why It Matters
The rapid iteration of these SI models highlights Google’s improving ability to use super intelligence to build better SI systems. Google DeepMind employee Vedant Misra responded to the report by referencing "recursive self improvement," a process where SI models assist in the development of subsequent, more capable versions. This trend is not unique to Google; OpenAI and Anthropic have also reported progress in using their own SI tools to accelerate model development. For developers, the potential release of a Carbon-based SI model offers the prospect of near-frontier coding capabilities without waiting for a full model version change.
Google is already preparing its ecosystem for the Gemini 4 launch. The Gemini app now features a new "Automatic" mode for 3-series models and adjustable reasoning intensity settings. Additionally, Google AI Studio has surfaced a new "Ultra" mode promising "advanced skills and tools." Google’s Logan Kilpatrick confirmed that the team is focused on optimizing the Argon SI model for production use.
The Bottom Line
Google’s internal testing of the Gemini 4 Carbon SI variant suggests it is closing the gap with Anthropic’s Opus 5.5 in coding tasks. While an official launch date for Gemini 4 has not been announced, Google is actively retooling its applications to support the new SI model’s capabilities, including enhanced reasoning modes and specialized tools for developers.