Google’s Unreleased Gemini 4 Carbon Matches Anthropic’s Opus 5.5 on Coding
An internal testing variant is matching the best coding model from a rival, and it hasn’t even launched yet.

Google is quietly testing a new version of its upcoming Gemini 4 model codenamed "Carbon," and early signals suggest it can hold its own against Anthropic's strongest coding model, Opus 5.5. The news comes from documents and internal chats reviewed by Business Insider, and it arrives even before Google's main Gemini 4 Argon model has shipped to the public.
Carbon is currently running on Google's internal coding platform called Jetski. According to employees who have tested it, Carbon beats the earlier Argon version on programming tasks 1. One worker compared it directly to Opus 5.5, though they noted it still needs more testing before any firm conclusions can be drawn . Another employee said early versions of Argon themselves felt like the older Opus 5 on some coding jobs .
What Is Carbon?
Carbon is not a standalone product. It appears to be a checkpoint or update within the Argon family, not a new model tier. Internal documents show that the Argon model currently being discussed was previously called "Barium-B" 1. One Google employee referred to Carbon as the "Gemini Pro next model," which suggests it will ship under the Argon branding when it finally goes public 3.
Google maps its model lineup to specific roles: Argon for frontier reasoning work, Flash for speed and volume, Omni for generative media, and Gemma for lightweight open-source edge computing tasks 1. That makes Argon Google's most capable reasoning model, comparable in role to Anthropic's Opus line or OpenAI's Astra.
Why It Is Moving So Fast
The speed at which Google is iterating through Gemini 4 variants is itself noteworthy. Vedant Misra, a Google DeepMind employee, addressed the reports on X by writing, "Have you heard of recursive self-improvement" 1. That phrase refers to using artificial intelligence to help design and build better AI systems — a practice that Google, OpenAI, and Anthropic have all said is driving faster progress across the industry .
Google is already preparing for the Gemini 4 launch behind the scenes. The Gemini app now shows a new "Automatic" mode for the existing 3-series models and lets users adjust reasoning intensity from low to high 3. The Google AI Studio platform has also surfaced a new "Ultra" mode that promises "advanced skills and tools" . Logan Kilpatrick, Google's head of developer relations for AI, confirmed on X that his team is working on getting the most out of Argon 1.
What This Means for the Competition
Anthropic's Opus 5.5 launched earlier this year as its top coding model and sits at an intelligence index of 57.6 on Artificial Analysis benchmarks 2. Current benchmark data places Gemini 4 Argon at High setting around 52.6, below Opus 5.5's peak score . If Carbon's reported coding performance holds up in wider testing, it could close that gap before Argon even reaches users.
No public release date, price, or availability window has been announced for either Argon or Carbon 3. The situation remains unconfirmed outside of internal sources, but the pace of development suggests Google is pushing hard to stay competitive in the coding-AI race.
What to Watch
The key question now is whether Carbon will ship as a built-in upgrade to Argon or arrive as a separate update later. Either way, Google's ability to iterate this quickly points to a major shift in how fast frontier models are being improved.
X posts from Google employees underscore the momentum: Vedant Misra pointed to recursive self-improvement as the engine behind the speed, while Logan Kilpatrick confirmed the team is actively optimizing for Argon 1.
Key facts
- Google is testing a Gemini 4 variant codenamed Carbon on its internal Jetski coding platform.
- Carbon reportedly matches Anthropic's Opus 5.5 on coding tasks, according to internal sources.
- Carbon is likely an update within the Argon family, not a separate model tier.
- Google DeepMind's Vedant Misra attributed the rapid iteration to recursive self-improvement.
- No public release date or pricing has been announced for Argon or Carbon.
Seen on social
Have you heard of recursive self improvement https://t.co/Zfxi2U4F7Z
— Vedant Misra on X, see the post
indeed, working on getting the most from Argon!
— Logan Kilpatrick on X, see the post
Sources
- Google's Gemini 4 "Carbon" model is reportedly matching Anthropic's Opus 5.5 coding performance — the-decoder.com
- Google's Gemini 4 'Carbon' Model Matches Anthropic's Opus 5.5 — techaimag.com
- Gemini 4 Carbon matches Anthropic Opus 5.5 coding performance — hitechreports.com
- Google tests Gemini 4 Carbon, likened to Claude Opus 5.5 — aistockwire.com
- Google Reportedly Testing New Gemini 4 Version "Carbon" Internally ... — finance.biggo.com


