Google’s new Gemini 4 Carbon model outpaces rivals like Claude Opus 5.5 and GPT-6 Astra in key benchmarks, achieving a top AA Index v5.0 score of 59. The blog post highlights superior performance in cybersecurity, coding, computer use, and knowledge work, noting it’s twice as likely to generate secure code as its predecessor Gemini 4 Argon. Carbon improves safety by reducing loopholes, enhancing transparency, and resisting prompt injection. Initially, the model will be limited to select cybersecurity defense firms for pre‑release testing, with broader access through Google’s Fairwind Program planned later, following the rollout path of Gemini 4 Argon.
Key Highlights:
- Highest AA Index v5.0 score (59) vs. Claude Opus 5.5 and GPT-6 Astra.
- Dominates cybersecurity, vulnerability detection, and secure coding.
- Safety enhancements: fewer shortcuts, better transparency, stronger prompt injection resistance.
- Release strategy mirrors Argon: phased rollout to trusted partners before public availability.
Key Entities:
{"companies": ["Google", "Anthropic", "Open AI"], "people": [], "locations": []}
Source: Engadget · Summarized by HeadlinesBriefing