Demis Hassabis touts Gemini 4 Carbon’s benchmark performance and safety research

What happened

Demis Hassabis says Gemini 4 Carbon is SOTA across key benchmarks and combines frontier capabilities with deep safety and security research.

Read next

• More changes in today’s briefingGoogle DeepMind announces Genie 4 for real-time world simulation with native audio

The story so far

Google DeepMind announced Gemini 4 Carbon with initial access limited to selected cyber defenders, while Artificial Analysis ranked it first on its Intelligence Index v5 with a score of 59; the model is not yet publicly available. Google said access would expand through its Fairwind Program to additional trusted partners as it gathers feedback and refines guardrails before broader availability for developers, businesses, and consumers. Artificial Analysis reported that Carbon led four of nine evaluations, but its coding-agent score using Antigravity CLI was 59, behind Claude Opus 5.5 at 62. Google reported cybersecurity scores of 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench, all measured without guardrails. Sundar Pichai said Carbon significantly improves on Argon in long-horizon tasks and has strengthened frontier safeguards. The restricted rollout drew criticism from Nathan Lambert, who objected to Google announcing benchmark scores for a model he could not publicly use. He later softened his criticism, saying he felt bad about the rant after hearing about difficulties getting a frontier model released through unofficial channels.