Google announces Gemini 4 Carbon, limited to cyber defenders at first

Google DeepMind announced Gemini 4 Carbon, which it calls its strongest model, with what it describes as state-of-the-art cybersecurity and long-horizon capabilities. The release comes two weeks after Gemini 4 Argon. Google says Carbon ranks first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59.

Carbon is not generally available. Starting this week, Google is giving it to a subset of cyber defenders and plans to phase in more trusted partners through its Fairwind Program. The company says it is taking part in the U.S. government's voluntary pre-release testing process and is still gathering feedback and iterating on guardrails before opening it to developers, businesses and consumers.

Google cites SEC-bench Pro (88.2%) for finding vulnerabilities, ExploitGym (54.8%) for exploitation and CWE-bench (71.0%) for patching. It says internal tests show more than twice Argon's success rate at generating secure code, and that the Chrome team found Carbon more than 10 times better than another lab's frontier model at producing high-quality patches. On WebDev Arena, Carbon gained 89 Elo points to reach fourth place at 1768.

Pricing matches Argon's introductory rate: $2 per million input tokens and $10 per million output tokens, with cached input tokens 95% cheaper. Google says Carbon is built for sustained agentic work and that internal teams have seen it run for days on multi-step engineering problems such as tracing and patching subtle bugs.

Ars Technica is more skeptical of the framing. It notes that Carbon beats Argon by slim margins on every test Google published, and that Google presents it as a new model although it could be a post-training upgrade of Argon. It also points out that every frontier model scores in the same neighborhood, with Carbon ahead of GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1 by only a few points.

The cybersecurity focus is the notable part. Google is positioning Carbon for the whole defense lifecycle of detection, validation and remediation, and is restricting early access to defenders. Dual-use capability, particularly in exploitation benchmarks, is the reason Google gives for a deliberate rollout. The Chrome and Wiz comments are Google-side claims and have not been independently verified in the coverage.

On safety, Ars Technica reports that in the Grey Swan evaluation Carbon resisted every attempted prompt-injection attack, a 0 percent success rate for attackers versus 0.7 percent for Argon, 1 percent for Opus 5.5 and 8.5 percent for GPT-6 Astra. The Verge adds that Google says Carbon halves unsupported claims compared with Argon.

Google has not promised a timeline for broad availability. Ars Technica says that if Carbon follows Argon, it could arrive in a couple of weeks, probably first for AI Ultra subscribers and Gemini API developers, though that is its expectation rather than a Google commitment.

The introductory price is temporary, and Ars Technica reports it will double once the period ends. Argon's discounted rate runs through the end of November, but Google has declined to give an end date for Carbon.