Google’s Gemini 4 Carbon went live on Vertex AI and was added to the Google Cloud SKU catalog on Nov 10 before an official general release announcement. Calls from a new Google Cloud project to the global endpoint using the “gemini-4-carbon” model slug returned responses identifying the model as Gemini 4 Carbon, even though it had not yet been added to the Gemini API model list or the Vertex catalog.

Google unveiled the frontier model on Oct 28, two weeks after Gemini 4 Argon, initially restricting access to vetted cybersecurity defenders in its Fairwind Program because of its vulnerability discovery and exploitation capabilities. In new cybersecurity benchmark results shared Nov 3, Carbon topped SEC-bench Pro at 88.2% for writing proofs of concept that trigger real V8 and SpiderMonkey vulnerabilities, beating OpenAI’s GPT-6 Astra at 85.4%, GPT-6.1 Terra at 83.3% and GPT-6.1 Sol at 78.8%.

Carbon earlier ranked No. 1 on Artificial Analysis’s Intelligence Index v5 with a score of 59, ahead of Anthropic’s Claude Opus 5.5 at 58 and GPT-6 Astra at 57, and topped Mercor’s APEX-Agents leaderboard with a 90.4% mean score across investment banking, management consulting and corporate law tasks. Standard pricing is set at $4 per million input tokens and $20 per million output tokens, with a 50% launch discount to $2 and $10. At standard rates, Carbon costs $7.76 per Intelligence Index task, 23% less than Claude Opus 5.5 at $10.04.