Google Releases Gemini 4 Carbon and Previews Agent Swarm

Google has made Gemini 4 Carbon, its most powerful model to date, available to all paid developers and AI Ultra subscribers, just two weeks after a targeted release to cybersecurity defenders through the Fairwind program. Carbon is positioned for extreme complex tasks such as multi-step coding, open-ended research, and cybersecurity, complementing the earlier Gemini 4 Argon, which remains the go-to for most software engineering and knowledge work.

Technically, Carbon ranks first on CursorBench 4.0, WANDR, and AutomationBench, with per-task cost less than half that of the next-best external model. Early tests show its Terminal-Bench cheating rate and prompt injection success rate on the Gray Swan benchmark both dropped to 0%, secure code writing is twice as fast as Argon, and it discovered 88% of vulnerabilities on Sec-Bench Pro. Pricing matches Argon at $2/$10 per million input/output tokens, with a 95% discount on cached input, though it consumes more tokens during reasoning and verification. Additionally, Google previewed a swarm feature on its Antigravity agent platform to enhance long-horizon agent collaboration.

This move signals Google's return to the front tier, competing with Anthropic and OpenAI. Carbon's breakthroughs in security and long-horizon tasks provide a reliable foundation for unattended agents, while the swarm feature points to a future of multi-agent collaboration, potentially reshaping how developers build complex AI applications.