Gemini 4 Carbon
Gemini 4 Carbon is Google’s model for long-horizon and technically demanding workflows. Google announced it October 28 and made it broadly available through selected products on November 10, 2026.
At a glance
| Documented milestone | Oct 28, 2026 |
|---|---|
| Context window | 1,048,576 tokens |
| Input / output | Text, image, video, audio and PDF input; text output |
| Open weights | Not verified |
| API text price | $2 input / $10 output per 1M tokens. Standard paid API rate; includes thinking tokens. No free tier. Introductory rate through November 30, 2026; $4 / $20 starting December 1. |
What changed
The distinction between announcement and release matters here: the initial rollout was limited to trusted partners; the later milestone opened developer and subscriber access. Google highlights software engineering and cybersecurity as intended uses, but its benchmark claims should be evaluated in context.
Who it's for
Developers evaluating complex, multi-step engineering work can test Carbon on their own tasks. Researchers and security teams should check eligibility and the safeguards that apply to their chosen access path.
How to try it
Google lists access through the paid Gemini API tier in AI Studio and Android Studio, as well as through selected Google AI subscription products. Availability and entitlements vary by product.
What it costs
Gemini API: $2 input / $10 output per 1M text tokens. Standard paid API rate; includes thinking tokens. No free tier. Introductory rate through November 30, 2026; $4 / $20 starting December 1.
Introductory pricing is time-limited; check the official pricing and the applicable rate before running workloads.
Documented events
Sources
- Google: Introducing Gemini 4 Carbon · checked Nov 11, 2026
- Google: Gemini 4 Carbon is now available · checked Nov 11, 2026
- Google AI Studio: Gemini API pricing · checked Nov 11, 2026
- Google AI Studio: Gemini 4 Carbon model · checked Nov 11, 2026
Prices and availability can change. We report vendor documentation; we have not independently benchmarked these models.