Google has opened Gemini 4 Carbon, the strongest model in its line-up, to every paying developer and enterprise, two weeks after giving it first to a group of cybersecurity defenders. The headline is the price: Carbon costs the same per token as Gemini 4 Argon during the introductory period, $2 per million input tokens and $10 per million output. For teams choosing between the two, the real question is not price per token but how many tokens each spends on a job.
Google is explicit about the division of labour. Argon remains its recommended model for most software engineering and knowledge work. Carbon is for the hardest tasks, complex multi-step coding and debugging or open-ended research, and it deliberately uses more tokens than Argon to explore, reason and verify. Google says it ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench at more than 2X lower cost per task than the next best external model, and second on a new Multimodal Knowledge Agent benchmark built with Artificial Analysis, at more than 6X lower average cost per task than the leader.
The launch comes with tooling. Google is previewing /swarm in Antigravity, which turns one prompt into a coordinated team of agents, with Carbon taking roles such as design, coordination and auditing while lighter models work in parallel workspaces and adversarial verifiers test each change. Google's examples include a 3D reconstruction of an office built from seven planning PDFs over 10 overnight loops. On security, Google says Carbon, accessed through its Fairwind Program, leads on benchmarks for finding, exploiting and patching vulnerabilities, and that in its internal Project Fortify it reached 100% recall on scanning and 95% precision on triage.
Access is broad. Carbon is in the Gemini API through AI Studio and Gemini for business, and rolling out to Google AI Ultra subscribers in Antigravity, the Gemini app and AI Mode in Search. For anyone comparing it with Claude Opus or OpenAI's top models in Cursor or Claude Code, the cost-per-task claims are the ones to check, because a model that spends twice the tokens at the same rate costs twice as much.
The caveats are familiar. Nearly every number is Google's, and several benchmarks were run or designed with partners. The introductory price does not last: according to our review of Carbon, it rises to $4 and $20 after 30 November. And Google itself says prompt-injection robustness still needs work, even after a reported 0% attack success rate on one benchmark.