Google Releases Gemini 4 Carbon, Its Most Powerful Model, and It Comes With a Countdown on Pricing

Google just put its heaviest AI model into people's hands. On Tuesday, it released Gemini 4 Carbon to developers, businesses, scientists, and paying consumers, calling it the strongest model it has ever built. It sits above Gemini 4 Argon, which Google says is still the better everyday choice for most coding and office work. Carbon is meant for the really hard stuff, and it spends more effort thinking and double-checking its answers to get there.

If you want to try it, developers can get it through Google AI Studio and Gemini for business. Google AI Ultra subscribers are getting it in the Gemini app, in AI Mode in Search and in Antigravity, Google's coding tool. There's a pricing catch, though. For now, it costs $2 per million input tokens and $10 per million output tokens. After November 30, that doubles to $4 and $20. So developers have a few weeks to experiment cheaply.

Google is leaning hard on benchmarks. It 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. The more interesting claims are in science and security. Google's AlphaProof Code agents used Carbon to help prove important security properties of the Signal messaging protocol, work Google says normally takes expert teams years. In internal security testing, it caught every vulnerability in its scans, and the security firm Wiz says it helped find brand-new flaws in two big software products in under two hours.

There's also a preview of a feature called /swarm, which turns one prompt into a whole team of AI agents working together. That lands one day after Anthropic showed off 1,000 agents building an operating system, so the agent race is clearly heating up.