Google released Gemini 4 Carbon to the general public on Tuesday as its most advanced AI model to date, introducing a promotional token pricing structure alongside a preview feature that coordinates multiple autonomous agents.
The system arrives two weeks after Google distributed the model to a restricted group of cybersecurity specialists through its Fairwind program. Koray Kavukcuoglu, vice president of research at Google DeepMind and chief AI architect, said in a company statement that the model demonstrates the accelerated pace with which the organization seeks to push the technological frontier while reinforcing its safeguards in parallel.
Google said that Gemini 4 Argon remains the optimal choice for routine software engineering and everyday knowledge tasks, while Carbon targets tasks that test model intelligence limits. The company said that Carbon consumes more tokens to analyze problems deeply through extended verification loops.
Carbon achieves specific scores in cybersecurity benchmarks
In benchmark evaluations, Carbon scored 88.2 percent on SEC-bench Pro for vulnerability detection, 54.8 percent on ExploitGym for exploitation, and 71 percent on CWE-bench for patching code, according to metrics published by Google and cited by BraveNewCoin and DiarioTI. The model achieved 59 points on the Artificial Analysis Intelligence Index v5.0, placing it one point ahead of Anthropic’s Claude Opus 5.5 at 58 points, as reported by WWWhatsnew.
WWWhatsnew and VentureBeat said that Carbon’s hallucination rate climbs to 11 percent, compared with 4 percent for Argon.
A coordinator distributes tasks among specialized agents operating within isolated, parallel workspaces, while adversarial verifier agents test changes and can be monitored in real time.
Pricing and ecosystem availability
Carbon launched with an introductory rate of two dollars per million input tokens and 10 dollars per million output tokens, which will increase to four dollars and 20 dollars when the promotional period ends on November 30, according to La Razón. The model supports a context window of up to one million tokens and is accessible through the API in Google AI Studio and Gemini para empresas.
