Gemini 4 Carbon Is Generally Available: $2 and $10 Per Million Tokens Until November 30

Published:

Google’s Gemini 4 Carbon became generally available on November 10, 2026, priced at $2 per million input tokens and $10 per million output tokens through November 30. On December 1 the standard rate doubles to $4 and $20, according to Google’s own pricing page. The model is not available on the free tier. Here is what it costs, who can use it, and what the price comparison with Anthropic and OpenAI leaves out.

Price per million tokens

  • Standard, through November 30: $2.00 input, $10.00 output, $0.10 for cached context.
  • Standard, from December 1: $4.00 input, $20.00 output, $0.20 cached.
  • Batch and Flex: half the standard rate through November 30 ($1.00, $5.00, $0.05 cached), doubling on December 1.
  • Priority: $3.60 input, $18.00 output and $0.18 cached, rising to $7.20, $36.00 and $0.36.

Google’s page says output pricing includes thinking tokens, so a request that reasons at length costs more than the visible answer suggests. Context caching is implicit only; explicit caching is not supported.

How it compares

VentureBeat’s launch report lists these standard rates: Claude Fable 5.2 at $8 input and $40 output, Claude Opus 5.5 at $4 and $20, and GPT-6 Astra at $10 and $50. Carbon’s introductory price is a quarter of Fable 5.2’s. From December 1 it matches Opus 5.5 and is half the Fable 5.2 price. Our guide to the Fable 5.2 price cut covers Anthropic’s rates in more detail.

As a simple illustration using only the published rates, a workload of 10 million input tokens and 2 million output tokens costs $40 on Carbon through November 30 ($20 for input plus $20 for output), $80 from December 1, and $160 on Fable 5.2 ($80 plus $80). It is arithmetic, not a benchmark, and it ignores cached tokens and thinking tokens.

Price is not the same as job cost

VentureBeat cites Artificial Analysis, which puts Carbon at a score of 60 on its Intelligence Index at high effort, against 62 for Fable 5.2 at high effort. On its index tasks, the estimated cost at high effort was $7.76 for Carbon and $9.08 for Fable, a much smaller gap than the token prices imply. Gadget Review adds that Carbon took 26.1 minutes per task on average against 16.9 for Fable at high effort, and that the two configurations are not directly comparable. VentureBeat quotes the point directly: “The bigger question for buyers is how many tokens, retries and human corrections it takes to finish the job.”

Results vary by test. On Terminal-Bench 4.0, VentureBeat reports Carbon at 65.2% against 61.6% for Fable’s best listed run, while on FrontierCode 2.0 Fable leads, 49.1% to 40.6%. Google’s own claims on CursorBench and the WANDR research benchmark are described by VentureBeat as unverified.

Who can use it

  • Developers: the paid Gemini API in supported countries, through AI Studio, plus Gemini for business. VentureBeat says Google has not published a country-by-country schedule.
  • Consumers: Google AI Ultra subscribers, in the Gemini app, AI Mode in Search and Antigravity. VentureBeat lists the U.S. prices as $99.99 a month for Ultra 5x and $199.99 for Ultra 20x.
  • Security teams: a gated cyber version remains in Google DeepMind’s Fairwind program. Our earlier report covers why Google restricted the model’s security capabilities.

Carbon is a proprietary, hosted model. VentureBeat reports no downloadable weights or self-hosting licence, and no word on Vertex AI availability or regions.

What to watch

  • The price doubles on December 1, so costs planned on the introductory rate will rise.
  • Test with your own prompts and effort settings before comparing vendors, since cost per finished task depends on retries and thinking length.
  • The Google pricing page does not state the context window. VentureBeat attributes the 1 million token window to Artificial Analysis.

For other Google pricing, see our report on Gemini Omni 1.2 Flash.

Sources

Image: “Google-Deep Mind headquarters in London, 6 Pancras Square” by Gciriani, CC BY-SA 4.0, via Wikimedia Commons.

Like our coverage? Add stepphase.com as a preferred source on Google to see more of our stories in Top Stories.

Saif Ali Tai
Saif Ali Taihttps://xplorely.com/
I'm Saif Ali Tai, a tech blogger and software engineer with 8+ years of experience building apps, and I run StepPhase as its editor. I cover technology news, AI tools and security stories that matter to everyday users. Articles here are drafted with AI assistance and checked against the sources linked in each post.

Related articles

Recent articles