Google expands access to Gemini 4 Carbon and previews Antigravity agent teams

Google says paying developers and enterprises can use Carbon through the Gemini API in Google AI Studio and Gemini for business.[2] Carbon is not available on the Gemini API’s Free tier.[4] Google is rolling it out to all Google AI Ultra subscribers in Antigravity, the Gemini app and AI Mode in Search.[2] Scientists and researchers can also apply for a chance to use Carbon through Antigravity’s highest usage tier, which Google says is free for six months.[5] The available routes differ by audience and product, so teams should check which offering matches their account and intended workflow.

Google describes Carbon as a model for frontier performance in long-horizon agentic workflows spanning complex knowledge work, software engineering and scientific reasoning.[2] The company says its teams have seen it run for days on complex engineering tasks, persist through dead ends, and work on bugs in large codebases and low-level systems.[1] Google says Carbon uses more tokens than Gemini 4 Argon to explore thoroughly, reason deeply and run extended verification loops.[2] Google also says the additional reasoning tokens increase both token use and latency per turn compared with Argon.[4] That trade-off makes workload complexity and run time relevant when estimating costs.

Google lists Carbon’s introductory API price at $2 per million input tokens and $10 per million output tokens, the same introductory price as Argon.[2] Google AI Studio lists those prices through Nov. 30, 2026, after which input tokens cost $4 and output tokens cost $20 per million.[4] The listed introductory rates therefore have a defined end date, a detail teams budgeting for sustained or high-volume use should include in their estimates. Google says Argon remains the ideal choice for most software engineering and knowledge work, while Carbon is intended for tasks that test a model’s limits.[2]

Google is previewing `/swarm` in Antigravity, a feature that turns one prompt into a coordinated team of agents.[2] Google says specialized agents plan, build and verify work in parallel.[5] Its guidance is to use Carbon for high-leverage coordinator, architecture, cybersecurity-auditing and adversarial-verification roles, while lighter, faster models handle parallel file edits and routine implementation.[5] For teams considering the workflow, this points to assigning models different roles rather than using Carbon for every task. Google says feedback from initial cyber defenders will inform guardrails before it expands access.[3]