Gemini 4 Carbon goes GA at $2 per million input tokens, undercutting Fable 5.2

Google DeepMind released Gemini 4 Carbon, a frontier multimodal reasoning model built on the Gemini 4 Argon architecture with a 1-million-token combined input and output context window. It is available to developers, businesses, consumers and Google AI Ultra subscribers. Introductory pricing starts at $2 per million input tokens. Medium coverage puts that at roughly 75% cheaper than Claude Fable 5.2, which launched a day earlier.

Google positions Carbon as its strongest model for long, complex tasks. It claims first place on CursorBench 4.0, WANDR and AutomationBench, at less than half the cost per task of the next-best outside model. Cursor confirmed Carbon tops CursorBench at 60.4% at high effort, a post Demis Hassabis retweeted. Perplexity CEO Aravind Srinivas rolled Carbon out to all Pro and Max subscribers, saying it is 2.5x cheaper than Fable 5.2 with similar capability on agentic research tasks.

The launch turns this week into a direct price-performance contest. Anthropic cut Fable 5.2 pricing by 20% to $8/$40, but Carbon's introductory rate is far lower. Third-party integrations from Cursor and Perplexity arrived within hours, a sign that Google has fixed the distribution lag that slowed past Gemini launches.

The key caveat is the word "introductory": Google has not said how long the price lasts. Practitioners also report better reasoning consistency during large-scale refactoring, but independent long-context evaluations are still early. Expect more coding-tool switches, and possibly a pricing response from Anthropic or OpenAI.