Google makes Gemini 4 Carbon generally available via API, Gemini app, AI Mode and Antigravity, at the same introductory price as Gemini 4 Argon

Google has made Gemini 4 Carbon, its top model, generally available about two weeks after limited access for cybersecurity defenders under the Fairwind Program. It is in the API (Google AI Studio and Gemini for business) for all paying developers and enterprises at $2 per million input tokens and $10 per million output tokens, the same introductory price as Gemini 4 Argon. It is rolling out to Google AI Ultra subscribers in Antigravity, the Gemini app and AI Mode in Search. A /swarm multi-agent feature is in preview in Antigravity, and Google launched a new Multimodal Knowledge Agent (MKA) benchmark with Artificial Analysis.

Developers and enterprises can now test Google's strongest model on production workloads at a price equal to the cheaper Argon tier, which makes the cost-versus-quality comparison with rival models easy to run. Teams doing complex coding, agentic or security work have a concrete reason to run their own trials this week. Consumer impact is narrower, since app and Search access is limited to Ultra subscribers and is still rolling out. The word "introductory" means the price may rise, so budgets should not assume it holds.

The "#1" rankings on CursorBench 4.0, WANDR and AutomationBench, the "2X lower cost per task" claim, and the Project Fortify figures (100% recall, 95% precision) are all Google's own reporting, and the Fortify numbers come from an internal benchmark. MKA is a new benchmark co-launched by Google, and "#2" on it says little about real workloads. "Strongest model" is a marketing superlative, and the cyber capability itself was already known from the earlier limited release.

1. 01Independent evaluations (Artificial Analysis, LMArena, third-party coding and agent tests) that reproduce or contradict the claimed #1 rankings and the cost-per-task advantage.
2. 02Whether the introductory price is raised or ended, and how rate limits, latency and reliability hold up as paying developers scale up usage.
3. 03Evidence of real adoption: named customers moving production workloads to Carbon, plus whether the Antigravity /swarm preview reaches general availability and works reliably.