Gemini 4 Carbon opens to developers at Argon’s price until November 30, then doubles

Two weeks after giving Carbon to cybersecurity defenders through its Fairwind Program (debut), Google is releasing it to paying developers and enterprises via the API, and to Google AI Ultra subscribers in the Gemini app, AI Mode in Search and Google Antigravity. Google pitches it as its strongest model for long-horizon agent work, cyber, complex coding and scientific reasoning, with Argon remaining the default for most tasks. The introductory price matches Argon at $2 per million input tokens and $10 per million output; on November 30 it doubles to $4 and $20.

The story is now who gets to run the most expensive kind of thinking, and at what price. Google’s cheaper Hydrogen model followed days later to handle execution (Hydrogen).

Carbon is explicitly token-hungry; Google says it uses more tokens than Argon to explore, reason and run verification loops. Pair that with the /swarm preview in Antigravity, which turns one prompt into a coordinated agent team, and you get a product for huge jobs, like Google’s multi-agent reconstruction of an office building from city planning PDFs across ten overnight loops. Pricing at Argon’s level for three weeks invites teams to wire Carbon into pipelines before the price doubles. Fair business; teams should budget for the post-November rate, the same introductory-pricing pattern Google used with Argon (Argon launch).

Google claims first place 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 agent benchmark co-launched with Artificial Analysis at more than 6x lower cost. These are vendor-reported, and one benchmark is new and Google-partnered.

Google reports Carbon passed 92% of a 375-task agentic honesty benchmark, up from 70% for Argon, declining to invent missing experimental data 98% of the time. Red-team per-attempt breach rate fell to 0.8% from 3.1%. For science users, refusing to fabricate data matters more than any leaderboard. Google also cites Wiz finding zero-day flaws in two major SaaS providers in under two hours, its own team proving two open-source bugs exploitable so maintainers could patch them, and a 40-fold average increase in homologs retrieved for hard protein families. Its free science program, which opened Argon to up to 10,000 researchers, now includes Carbon.

A model with Google’s strongest cyber skills is now in many more hands, and Google says prompt-injection robustness still needs work. Keeping the cyber-specialized variant behind Fairwind is right, but misuse in the general release should be tracked and reported over the coming months, not only measured before launch, given Gemini’s own history in security testing (Gemini breaches). Google competing on cost per completed task pressures pricing across the frontier. More in our Google Gemini coverage.