AI Model Launch: Google Unveils Gemini 4 Carbon

Google released Gemini 4 Carbon on Wednesday. The new frontier AI model targets coding, agentic tasks, and cybersecurity improvements over previous versions.

Carbon now holds the top spot on the Artificial Analysis Intelligence Index v5.0. Claude Opus 5.5 and GPT-6 Astra currently rank behind it. The model also outperforms leading systems from OpenAI and Anthropic on several other benchmarks.

The release builds on Argon, a model Google introduced earlier this month. Access is currently restricted to a select group of cybersecurity defense companies. Google plans to widen availability to additional partners in its Fairwind Program over time.

The company stated the release reflects the pace at which it is pushing the frontier forward. Google described a significant leap in cyberdefense capabilities, noting the model is better at finding vulnerabilities and generating secure code. Carbon ranks first on SEC-bench Pro, which measures the ability to discover real-world bugs. It also leads on ExploitGym, a test for turning vulnerabilities into working exploits.

Google says Carbon is designed for longer tasks. Internal teams have observed the model running for days to persist through dead ends on complex engineering tasks. The company adds that the model is better at avoiding shortcuts, resisting prompt injection attacks, and collaborating with users.

A focus on transparency is part of the new design. Google says Carbon produces unsupported claims at half the rate of Argon. The model will launch at the same introductory rate as Argon: $2 per million input tokens and $10 per million output tokens.