AI

Gemini 4 Carbon opens to everyone: price doubles on December 1

Google's Gemini 4 Carbon is out of limited access. What it costs now, when the price doubles and what the model has already built in Antigravity.

Google has opened Gemini 4 Carbon to everyone. Two weeks ago the model was announced only for “a subset of cyber defenders”, and on November 10 it became available to everyone else. Developer Simon Willison, author of a well-known blog on language models, calls Carbon Google’s best Gemini ever.

At launch the model costs $2 per million input tokens and $10 per million output tokens. On December 1 the price doubles to $4 and $20. There are three thinking levels, low, medium and high, with medium as the default.

Willison tested the model in Antigravity, Google’s AI development environment. He asked for an animation of a giant mech from Yonkers standing up and stomping across Manhattan, Queens and Brooklyn using real OpenStreetMap data. He had given the same task to Claude Fable 5.2 the day before. A finished file with no external dependencies came only after a separate request to build everything into one standalone HTML file.

Another example appears in a YouTube video. Using the /swarm feature in Antigravity, Carbon built an open-world 3D game where the player explores Google’s Mountain View campus as a pelican riding a GBike. The model coordinated subagents that made 3D assets in Blender, wrote the browser game in Three.js with WebGL shaders and added sound effects and Lyria music.

Carbon’s launch price matches GPT-6.1 Sol and Claude Sonnet 5.5, and after December 1 it will match Claude Opus 5.5. The model was already known from benchmarks: KRAKOBIT reported that Gemini 4 Carbon tops Claude Opus 5.5 in benchmarks and hallucinates less, and that it leads vulnerability benchmarks.

Source: Simon Willison