Google has announced Gemini 4 Carbon, its newest top-tier AI model, Ars Technica, CNET and The Verge report. The company unveiled it on Wednesday, just two weeks after it released Gemini 4 Argon.

Google says Carbon makes a "significant leap" in cybersecurity. The company says it is better at finding software flaws, checking them and fixing them. Ars Technica reports Google says Carbon is twice as likely as Argon to write code free of security holes. The Verge reports Carbon holds the top spot on SEC-bench Pro, a test of how well an AI finds real-world bugs.

Google says Carbon beats rival models from OpenAI and Anthropic on most of its tests, including GPT-6 Astra and Claude Opus 5.5. Ars Technica notes that the gaps are just a few points and that top models all score in a similar range. Google also says Carbon can work on long tasks for days and makes half as many unsupported claims as Argon, The Verge reports.

Most people cannot use it yet. For now, only select cybersecurity companies have access, the outlets report. The Verge says it will later expand to other partners. CNET reports Google would not say when wider access will open, and that Google says it is taking part in a US government review process. The model will cost developers $2 per million input tokens and $10 per million output tokens, the same as Argon.

Google also unveiled Genie 4 this week, CNET reports. It is a world model that builds simulated worlds and can now add sound, such as dialogue and background noise. It is not yet public either, but developers can join a waitlist.

Argon was Google's first new top model in 11 months, CNET reports.

What we know: Google has announced Gemini 4 Carbon and says it leads on cybersecurity tests, with access limited for now.
What we don't: When the public will be able to use Carbon or Genie 4.