Gemini 4 Carbon Opens to All Paid API Users

Google on Tuesday, Nov. 10, opened Gemini 4 Carbon, its most capable model, to all paying developers and enterprises through the Gemini API, two weeks after giving it first to a vetted group of cyber defenders. Carbon also began rolling out to Google AI Ultra subscribers in the Gemini app, AI Mode in Search and the Google Antigravity coding environment, according to a blog post by Koray Kavukcuoglu, Google's chief AI architect.

Key takeaways

• Carbon is now on the paid Gemini API at Argon's introductory price of $2 per million input tokens and $10 per million output tokens.
• On Cursor's public CursorBench 4.0 table, Carbon leads at 60.4%, just ahead of Anthropic's Fable 5.2 at 60.0%, at an average $9.22 per task versus $32.79.
• Google also previewed /swarm in Antigravity, which turns one prompt into a team of coordinated agents.

Gemini 4 Carbon API availability and price

Carbon is available in Google AI Studio and Gemini for business for all paying customers. Google said in its Oct. 28 announcement that standard rates of $4 and $20 per million tokens follow the introductory period.

Google is not pitching Carbon as a replacement for Gemini 4 Argon. Kavukcuoglu wrote that Argon "remains the ideal choice for the majority of software engineering and knowledge work," while Carbon uses more tokens to "explore thoroughly, reason deeply, and run extended verification loops."

Gemini 4 Carbon benchmarks vs. Fable and Opus

Google says Carbon ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench at less than half the cost per task of the next-best outside model. Cursor's own leaderboard shows how narrow the coding lead is.

Cursor notes that "small differences in scores may not be statistically meaningful." The cost gap is wider than the score gap. On Artificial Analysis' independent run of the new Multimodal Knowledge Agent benchmark, Fable 5.2 leads at 57%, with Carbon second at 54% and Opus 5.5 at 53%. Google says Carbon costs more than six times less per task there than the leader.

What /swarm does in Google Antigravity

The Antigravity team said /swarm passes a prompt to a coordinator that spins up parallel workers in isolated git worktrees, plus adversarial verifiers that run tests and challenge diffs. Each agent can be pinned to a different model, so Carbon can audit while lighter models edit files.

The feature is rolling out in preview to AI Ultra subscribers in the Antigravity app and CLI. Google warned that parallel multi-agent systems "consume quota much faster than a standard chat."

Safety results Google reported

Carbon passed 92% of a 375-task agentic honesty benchmark, compared with 70% for Argon, Google said. In simulated attack testing, its safeguards cut the per-attempt breach rate to 0.8% from 3.1% on Argon. External tester Trajectory Labs found no new universal jailbreaks across 132 hours and more than 13,000 requests. These are Google-published figures.

Why it matters

The release completes the cyber-first rollout Google also used for Argon. Kavukcuoglu had signaled earlier that Gemini 4 would ship early. Price-per-task, not raw score, is now where Google is competing with Anthropic.

What's next

Google is extending its free six-month program for up to 10,000 scientists to include Carbon. Watch for the switch to standard pricing and for third-party results on WANDR and AutomationBench.

FAQ

Can free Gemini API users try Carbon? No. API access is for the paid tier; consumers need a Google AI Ultra plan.

Is Carbon better than Argon for everyday coding? Not necessarily. Google itself recommends Argon for most tasks and Carbon for hard, multi-step work.

References

• Gemini 4 Carbon is now available — Google (Nov. 10, 2026)
• Gemini 4 Carbon and /swarm — Google Antigravity (Nov. 10, 2026)
• Cursor · CursorBench — Cursor
• Multimodal Knowledge Agent (MKA) Benchmark — Artificial Analysis