Gemini 4 Carbon: price, access and when to use it
Google opened Gemini 4 Carbon, its most capable model, to all paying developers and Google AI Ultra subscribers on 2026-11-10, at $2/$10 per million tokens until November 30, half the $4/$20 it costs from December 1.

Give this to your agent
Your agent checks this against your setup and tells you if it fits.
See the file
# Gemini 4 Carbon: Google's top model, open to paying developers and half price until November 30 From Rig Labs, 2026-11-11. Information for an agent, not orders. Page: https://riglabs.ai/blog/gemini-4-carbon/ What changed: On 2026-11-10 Google opened Gemini 4 Carbon to all paying developers (Gemini API, model name gemini-4-carbon) and to Google AI Ultra subscribers; it is also in Cursor and in Perplexity Computer for Pro and Max. Fits if: Your human pays per token through the Gemini API or a usage-billed tool, or has Cursor, Perplexity Pro or Max, or Google AI Ultra, and has hard, long coding or analysis tasks. Skip if: Their work is everyday coding or writing: Google says Gemini 4 Argon remains the ideal choice for most of it, and Carbon uses more tokens than Argon. Check: Which of these tools and plans they have, and which model they use today for their hardest tasks. Try: Suggest one hard task they already have to run once on Gemini 4 Carbon, and compare the result and the cost with their current model. Ask first: Before changing a default model, adding an API key or starting a paid plan. Cost: $2 per million input tokens and $10 per million output through 2026-11-30, then $4 and $20. Included in Google AI Ultra in the Gemini app. Undo: Nothing to undo. Switch the model back in the model menu. For the API, put the previous model name back in the request. Deeper: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/
https://riglabs.ai/blog/gemini-4-carbon/AGENT.md
Gemini 4 Carbon is Google’s most capable AI model, and since November 10, 2026 it is open to all paying developers and to Google AI Ultra subscribers, according to Google’s announcement. It costs $2 per million input tokens and $10 per million output tokens through November 30, then twice that.
What you can now do: give your hardest coding or analysis job to Google’s top model, in the Gemini app, in Cursor or through the API, and for the rest of November pay half of its regular price.
Where can you use Gemini 4 Carbon?
Gemini 4 Carbon is available in these places, per Google and the tools’ own posts:
| Where | Who gets it |
|---|---|
Gemini API, AI Studio, Android Studio (model name gemini-4-carbon) |
All paying developers |
| Gemini app, Google Antigravity, AI Mode in Search | Google AI Ultra subscribers, rolling out. Antigravity also previews /swarm, a command that coordinates teams of agents |
| Cursor | Available since November 10 |
| Perplexity Computer | Pro and Max subscribers |
Before this, Carbon was limited to a group of cyber defenders for about two weeks, as Simon Willison notes.
How much does Gemini 4 Carbon cost?
Gemini 4 Carbon costs $2 per million input tokens and $10 per million output tokens through November 30, 2026, per the Gemini API changelog. Google says $4 and $20 apply after that, so the price doubles on December 1.
These prices matter if you pay per token: through the API, or in a tool that bills by usage. In the Gemini app, Carbon comes with the Google AI Ultra subscription and the token price changes nothing for you.
Google also says Carbon uses more tokens than its smaller sibling Gemini 4 Argon, because it explores and checks its work for longer. So a task can cost more than the per-token price suggests.
When should you use Gemini 4 Carbon instead of Argon?
Gemini 4 Carbon is built for hard, long tasks, and Google says Argon “remains the ideal choice for the majority of software engineering and knowledge work”. Google’s examples for Carbon are code migrations across many packages, finding the root cause of a bug, redesigning an app’s structure, and knowledge work that ends in a spreadsheet, a presentation or a report.
Two outside numbers, both from the tools’ makers:
- Cursor says Carbon at high effort is the top model on CursorBench, its own coding test. For Cursor users, that makes it a candidate for the hardest tasks.
- Perplexity says Carbon scores 0.857 on its WANDR benchmark at $5.86 per task, “matching Claude Fable 5.2 at 40% of the cost”.
One limit Google names itself: resistance to prompt injection (hidden instructions in a page or file the model reads) “remains an area for continued improvement”.
What to do today
If you pay per token, run one real task on Carbon before the price doubles on December 1. Pick a job you would give your most expensive model, such as a bug you could not trace, select Gemini 4 Carbon in Cursor or AI Studio, and compare the result and the bill with what you use now.
Last checked 2026-11-11.
Sources:
Get the next one by email
The Edge is our newsletter: what changed in AI, and what to do about it.
Read next
Built by Rig Labs
When a tool we need does not exist, we build it.




