News
Google says Gemini 4 Carbon is its next frontier model, with benchmark gains in cybersecurity, coding and long-horizon tasks.
Google said on Oct. 28, 2026 that Gemini 4 Carbon is its next frontier model, positioning it as the company’s strongest system to date with a focus on cybersecurity defense and long-horizon tasks. The announcement said Carbon ranks No. 1 on the updated Artificial Analysis Intelligence Index v5.0 at 59 and outperforms prior Gemini models on several coding, agentic and security benchmarks.
Evidence and context
In Google’s post, Carbon is described as delivering a state-of-the-art 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation and 71.0% on CWE-bench for patching. The company also said internal testing showed more than a 2x uplift in secure-code generation versus Gemini 4 Argon, and that high-severity misalignment alerts in live coding environments with internal teams fell by almost 80%.
Google said Carbon is built for sustained agentic workflows and long-running engineering work, including multi-step bug root cause analysis and patching across large codebases. The company said it improved web development output as well, with Carbon rising 89 Elo points to No. 4 at 1769 on WebDev Arena, according to the post.
The announcement also included pricing, with Carbon set to launch at the same introductory rate as Argon: $2 per million input tokens and $10 per million output tokens, while cached input tokens are priced at 95% off the input token price. Google said the model will first be made available to a subset of cyber defenders and then gradually extended to additional trusted partners through its Fairwind Program before a broader release to developers, businesses and consumers.
Google said Carbon builds on the same multi-layered safety stack as Argon, including prompt-injection resilience and chain-of-thought monitoring. The post says Carbon reduced successful cheating on Terminal-Bench 4.0 to 0.0% and achieved 0.0% attack success on the Gray Swan indirect prompt injection benchmark in internal testing.
The company did not publish a public model card or technical specification sheet in the material provided, and the available background notes say verified Carbon-specific architecture details, context window, parameter count and release timing remain unconfirmed outside the Oct. 28 announcement. Google framed the rollout as deliberate and said it is engaged in the U.S. government’s voluntary pre-release model testing process.
The timing of a wider public launch remains unknown from the material provided.
