What matters in AI.

Subscribe

Google makes Gemini 4 Carbon open to all paying developers

Google tells that Carbon is first on 3 benchmarks at more than 2 times lower cost than the next external model.

Google made Gemini 4 Carbon available today to developers, enterprises, scientists and consumers. Google gave Carbon to cyber defenders in its Fairwind Program 14 days before. Carbon follows Gemini 4 Argon. The price is the same as for Argon: $2 for each 1,000,000 input tokens (small pieces of text) and $10 for each 1,000,000 output tokens.

How it works

Carbon uses more tokens than Argon for more checks of its work. Google tells that Argon stays the tool for most software and knowledge work. In Google Antigravity, a preview tool, /swarm, turns one instruction into a team of agents.

The numbers

Google tells that Carbon is first on CursorBench 4.0, WANDR and AutomationBench, at more than 2 times lower cost for each task. On a new multimodal test it is 2nd, at 6 times lower cost than the top model. In Project Fortify, an internal Google test, it got 100% recall on scanning and 95% precision on triage.

What is not known

The cyber scores come from the Fairwind Program. Google gives no data on the bugs it found. The cost numbers come from Google and its partners. Carbon uses more tokens than Argon, thus the cost for each task can be different from the price for each token.

Insight: Google puts the cost of a finished task, not the top score

On the GDPx test of SemiAnalysis, Carbon got 10.9% and Claude Opus 5.5 got 12%. A task cost $16.29 with Carbon and $51.41 with the Anthropic model. This reads as a price play: Google puts the cost for each task first, not the top score. Leaders must compare the cost of each task on their work, not one score.

Sources

  1. Gemini 4 Carbon is now availabledeepmind.google
  2. Gemini 4 Carbon is now availableblog.google
AI MATTER · NEWS · AI MATTER · NEWS ·10 NOV2026

Posted

Tags

Learn the terms in this story

More in Models

All Models news