Gemini 4 Carbon: Our next frontier model
Web page · https://deepmind.google/blog/gemini-4-carbon-our-next-frontier-model
First seen by Oriel on .
AI summary
Google announces Gemini 4 Carbon, a frontier model focused on long-horizon agentic work and cybersecurity, reporting leading benchmark results and a cautious rollout to cyber defenders.
- Carbon ranks first on Artificial Analysis Intelligence Index v5.0 with a score of 59, and reaches #4 on WebDev Arena after gaining 89 Elo points.
- It achieves 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench for vulnerability discovery, exploitation, and patching.
- The model is designed for sustained workflows, reportedly running for days while debugging and patching complex issues across large codebases and low-level systems.
- Carbon reduced Terminal-Bench 4.0 cheating to 0.0%, unsupported claims by approximately 2x, overreach by 26%, and under-service by 29% over Argon.
- Access begins with selected cyber defenders through the Fairwind Program; the introductory price is $2 per million input tokens and $10 per million output tokens.
#gemini #cybersecurity #ai agents #software engineering
Generated by gpt-5.6-sol. Check the original for details.
Aliases
| Source | External ID | URL |
|---|---|---|
| Google DeepMind Blog | https://deepmind.google/blog/gemini-4-carbon-our-next-frontier-model/ | https://deepmind.google/blog/gemini-4-carbon-our-next-frontier-model/ Read discussion |