Google

AI · Generative AI · Developer Tools

Launch

Google launches Gemini 4 Carbon, its strongest AI model yet

It launched a day after Anthropic's newest model reclaimed the top benchmark spot, yet undercuts rivals on price by four to five times.

• Google began rolling out Gemini 4 Carbon on November 10, 2026, to paying developers, enterprises and Google AI Ultra subscribers, two weeks after giving it first to cybersecurity defenders through its Fairwind Program.
• Carbon is built for long-horizon agentic work like multi-step coding, debugging and open-ended scientific research, while Google's prior model Argon remains the default for everyday software and knowledge tasks.
• Carbon launches at an introductory $2 per million input tokens and $10 per million output tokens, the same price as Argon and roughly a fifth of OpenAI's GPT-6 Astra and a quarter of Anthropic's Claude Fable 5.2.
• Google is previewing a '/swarm' feature in its Antigravity coding app that turns a single prompt into a coordinated team of AI agents, with Carbon handling planning and auditing while lighter models do routine edits.
• Google says Carbon ranks #1 on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench benchmarks while costing more than twice as little per task as the next-best external model.
• The release lands a day after Anthropic's Claude Fable 5.2 reportedly knocked Carbon off the top of a leading independent model ranking, underscoring how quickly frontier benchmark leads are changing hands.
• With two frontier model launches in barely a month, Google is signaling that cost-per-task and agentic reliability, not just raw benchmark scores, are becoming the real battleground among AI labs.