Google ships Gemini 4 Carbon, tops AA Intelligence Index at 59
TL;DR
- Carbon tops Intelligence Index v5 at score 59, one point ahead of Claude Opus 5.5, but independent benchmarking confirms it still trails Opus 5.5 specifically on agentic coding tasks.
- Google's introductory price of $2/$10 per million tokens doubles to $4/$20 post-launch, narrowing the cost advantage being marketed against rivals over time.
- Carbon's hallucination rate jumped to 11% from Argon's 4%, a material reliability gap in the cyber defense deployments Google is prioritizing first.
Google released Gemini 4 Carbon on October 28 and placed it at the top of the Artificial Analysis Intelligence Index v5.0 at a score of 59, framing the launch around cybersecurity rather than general-purpose capability. First access is routed through the company's Fairwind Program to 'a subset of cyber defenders' before developers and consumers; in the announcement post, Google DeepMind SVP Koray Kavukcuoglu calls it 'our strongest model with state-of-the-art cybersecurity and long horizon capabilities.'
The pricing is aggressive on the way in. Introductory rates are $2 per million input tokens and $10 per million output, with cached input tokens priced 95% off; the step-up to $4 and $20 after the introductory window is already published. On the security evaluations Google led with, Carbon posts 88.2% on SEC-bench Pro for vulnerability detection, 71.0% on CWE-bench for patching, and 54.8% on ExploitGym, and records a 0.0% attack-success rate on Gray Swan's Indirect Prompt Injection Benchmark.
The sharpest internal claim comes from Chrome. Carbon 'performed more than 10X better than another lab's large frontier model when generating high quality patches,' Google says of the Chrome team's own tests, and reports 'more than 2x uplift in success rate for generating secure code over Argon,' the previous model.
None of those headline numbers have been reproduced outside Google. The 10x Chrome comparison does not name the rival model, the Gray Swan 0.0% is Google's own count against an evaluated attack set, and Fairwind's membership is not disclosed.
The launch lands the same day Anthropic's Frontier Red Team said an unreleased Claude chained flaws into working browser and OS exploits; cyber-defense capability is where the labs are currently fighting out the frontier.
What others are reporting
-
The Decoder Read →
Triangulates Carbon's score across multiple independent indices; confirms Carbon still trails Claude Opus 5.5 on agentic coding despite the overall top rank.
Carbon is built for multi-day, autonomous development workflows designed for much harder, long-horizon engineering problems.
-
Latent Space Read →
Raises hallucination risk as a specific concern for cybersecurity use cases; reports Google runs overnight agent loops and swarms internally as standard pre-launch workflow.
Carbon first goes to trusted cyber defenders through the Fairwind Program, with wider developer access promised as soon as possible.
-
Fello AI Read →
Surfaces the full pricing ladder including post-introductory doubling and the hallucination rate jump from Argon 4% to Carbon 11%; estimates mid-November for broader public release.
Carbon ranks first on the Artificial Analysis Intelligence Index v5.0 with a score of 59, one point ahead of Claude Opus 5.5.
-
Blockchain.News Read →
Focuses on the deliberate rollout sequence and U.S. government voluntary pre-release testing as a procurement positioning move rather than a standard safety step.
Carbon ranks No. 1 on the updated Artificial Analysis Intelligence Index v5.0 at 59 and outperforms prior Gemini models on coding, agentic and security benchmarks.
Shared on Bluesky by 1 AI expert
Originally reported by blog.google
Read the original article →Original headline: Google Launches Gemini 4 Carbon, Takes #1 on Intelligence Index v5 at Score 59 With Cyber Defense Focus