Google Deepmind releases Gemini 4 Carbon just two weeks after Argon with gains across the board
decoder ·
Google DeepMind has released Gemini 4 Carbon, its latest frontier model, just two weeks after launching Gemini 4 Argon. The new model tops the Artificial Analysis Intelligence Index v5 with a score of 59, significantly outperforming its predecessor's 50 points. Designed for complex, multi-day autonomous development workflows, Carbon demonstrates substantial improvements in coding, scientific reasoning, and cybersecurity capabilities. Independent evaluations from platforms like Vals AI and Arena.ai place it second or first across various benchmarks, often offering better value than competitors like Claude Opus 5.5. Notably, Carbon leads in security-specific tests, including vulnerability discovery and patch writing, prompting Google to restrict initial access to cybersecurity experts under a voluntary U.S. government pre-testing process. Despite higher hallucination rates compared to Argon, the model maintains competitive pricing at $2 per million input tokens and $10 per million output tokens. This strategic rollout ensures that advanced safety guardrails are refined before broader public availability.
decoder