Google Releases Gemini 4 Carbon at $2 per Million Tokens, Then Doubles the Price After November 30

A day after Anthropic shipped its new top model, Google opened up its own. On Tuesday it made Gemini 4 Carbon available to all paying developers and businesses, two weeks after giving it only to a group of cybersecurity defenders. It is also rolling out to Google AI Ultra subscribers in the Gemini app, AI Mode in Search and the Antigravity coding tool.

The opening price is aggressive. Carbon costs $2 per million input tokens and $10 per million output tokens, the same as the cheaper Gemini 4 Argon. That rate is introductory, though. From December 1 the public model costs $4 and $20, still far below the $8 and $40 Anthropic now charges for Claude Fable 5.2.

Built to burn more tokens

Google is not pitching Carbon as a replacement for Argon. Argon stays the default for most coding and office work. Carbon is meant for the hardest jobs, like long debugging sessions and open research, and it deliberately uses more tokens than Argon to explore, reason and check its own work. A low price per token can still add up to a large bill per task.

Google's answer is cost per task. 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. On a new document analysis benchmark built with Artificial Analysis, it comes second at less than a sixth of the leader's cost per task. These are rankings Google chose to publish, and outside testing will take a few weeks.

Alongside the model, Google is previewing a /swarm command in Antigravity that turns one prompt into a team of agents. Carbon takes the design and review roles while lighter models do the parallel work.

A cyber model with guardrails

Security is the other headline. Through Google's vetted Fairwind Program, a less restricted "Carbon for cyber" leads benchmarks for finding, exploiting and patching vulnerabilities. Google says Wiz researchers used it to find zero day flaws in two major SaaS providers in under two hours.

The public version is locked down harder than Argon. In Google's red team tests, its safeguards cut the success rate of a single attack attempt to 0.8%, from 3.1% on Argon. Google also claims a 0% success rate for indirect prompt injection on the Gray Swan benchmark, while admitting that its own automated attacks still find ways through in some cases.

Scientists in Google's free research program, which offers six months of top tier Antigravity access, are getting Carbon as well. For everyone else, the cheap window closes in three weeks.

Some offers on this page may be paid placements or contain affiliate links.