Google DeepMind releases Gemini 4 Carbon two weeks after Argon
Original title: Google Deepmind releases Gemini 4 Carbon just two weeks after Argon with gains across the board
Carbon leads Artificial Analysis Intelligence Index v5.0 at 59 versus Argon’s 50, with stronger security-benchmark scores.
Google DeepMind released Gemini 4 Carbon two weeks after Gemini 4 Argon. Google says Carbon is built for multi-day agentic coding and leads 15 of 16 of its selected tests, including long-context and security tasks run without safety guardrails. Independent indexes place it first on Artificial Analysis Intelligence Index v5.0 at 59, second on Vals AI at 70.36% accuracy, and second in Arena.ai’s Agent Arena. Google is rolling the model out gradually and taking part in U.S. government pre-release testing because of its security capabilities.
Unconfirmed: Several security and coding gains are Google’s own or internal results and are not independently verified.
- On Gray Swan’s indirect prompt-injection test, Google reports attack success fell from 0.7% to 0.0%.
- SEC-bench Pro 88.2%, ExploitGym 54.8%, and CWE-Bench fixes 71.0%, per Google.
- Answer accuracy rose from 48% to 58% while hallucination rose from 4% to 11%.
- Agentic coding on Antigravity still trails Claude Code with Opus 5.5 by three points.
- Chrome-team and in-house secure-code gains are internal results Google says cannot be independently verified.
- Vendors
- Google DeepMind
- Organizations
- Artificial AnalysisVals AIArena.ai
- Countries
- United States
- AI models
- Gemini 4 CarbonGemini 4 Argon
From the original article
5 min read at the-decoder.com
Just two weeks after Gemini 4 Argon, Google DeepMind is releasing Gemini 4 Carbon, which tops the Artificial Analysis Intelligence Index v5 with 59 points. Google says the model is built for extended agent workflows and the full cyber defense lifecycle. Because of those advanced security capabilities, access is limited to selected security experts at first.
By Matthias Bastian · the-decoder.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.