Google revealed Gemini 4 Carbon, a new frontier model that the company said led intelligence rankings; it was also reported to have lower hallucination rates in benchmark tests. *(Nov. 3)

Google announced Gemini 4 Carbon on Oct. 28, calling it its strongest model for long-running engineering work and cybersecurity; a Nov. 3 report highlighted its benchmark lead. Artificial Analysis’ Intelligence Index v5 scored Carbon 59, ahead of Anthropic’s Claude Opus 5.5 at 58 and OpenAI’s GPT-6 Astra at 57. Carbon led four of the index’s nine evaluations. On AA-Omniscience, it scored 54, with 58% accuracy and an 11% hallucination rate—lower than Opus 5.5 and Astra, both reported at 50%, though Gemini 4 Argon hallucinated less, at 4%. Google also reported 88.2% on SEC-bench Pro for vulnerability discovery and 71% on CWE-bench for patching. Initially, Carbon is limited to U.S. government testers and selected cyber defenders through the Fairwind Program; broader access has no announced date. Google says introductory API pricing will be $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20.