Google has introduced Gemini 4 Carbon, its latest frontier AI model, and says it delivers significant improvements in coding, agentic tasks and cybersecurity. The company announced the release on Wednesday.
According to Google, Carbon now holds the top position on the Artificial Analysis Intelligence Index (v5.0). Claude Opus 5.5 and GPT-6 Astra rank behind it. The model also outperforms leading systems from OpenAI and Anthropic on some other benchmarks.
Carbon builds on Argon, a model Google released earlier this month. For now, access is restricted to a select group of cybersecurity defense companies. Google plans to widen availability over time to additional partners in its Fairwind Program.
The company described the release as a reflection of the "pace at which we're pushing the frontier forward." It pointed to a "significant leap" in cyberdefense, saying the model is better at finding vulnerabilities and at generating secure code.
Two security-focused benchmarks back up that claim, according to Google. Carbon ranks first on SEC-bench Pro, which measures how well an AI model can discover real-world bugs. It also leads on ExploitGym, which tests whether a model can turn a vulnerability into a working exploit.
Google also says Carbon is designed for longer tasks. Its internal teams have watched the model "run for days to persist through dead ends to tackle complex, multi-step engineering tasks." The company adds that the model is better at avoiding shortcuts, resisting prompt injection attacks and collaborating with users.
Another stated focus is "transparency and honesty." Google says Carbon produces "unsupported claims" at half the rate of Argon.
On pricing, Carbon will launch at the same introductory rate as Argon: $2 per million input tokens and $10 per million output tokens. Google describes that price as competitive.
