Google's Carbon model tops one index, but access is still narrow

Gemini 4 Carbon leads an independent intelligence index and is priced like its predecessor, yet general release has no date and most safety claims are Google's own.

Google DeepMind announced Gemini 4 Carbon on October 28, 2026, calling it its strongest model. The announcement was written by Koray Kavukcuoglu, Google's Chief AI Architect. Carbon ranks first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59. It arrives two weeks after Gemini 4 Argon.

The headline results are in security and coding. Carbon scores 88.2% on SEC-bench Pro (finding vulnerabilities), 54.8% on ExploitGym (exploitation) and 71.0% on CWE-bench (patching). It rises 89 Elo points to fourth place on WebDev Arena, at 1768. Google says internal teams watched it work for days on multi-step engineering tasks, such as tracing bugs through large codebases. It also reports that Carbon more than doubles Argon's success at writing secure code, and roughly halves unsupported claims.

For abundance, the relevant fact is price. Launch pricing matches Argon: $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper. If a stronger model costs no more, then tutoring, research and coding help get better at the same price. That is the useful direction. But the price is described as introductory, so it may not hold.

The caveats are large. Access starts this week only for a subset of cyber defenders, then widens to trusted partners through the Fairwind Program. Developers, businesses and consumers get it "as soon as possible", with no date. The cybersecurity scores were run without guardrails. Most safety, misalignment and secure-code figures are Google's internal tests, not independent checks. The Chrome team's comparison names no rival model. Watch for the general release date, final pricing, and outside evaluations.

Source: Google DeepMind