Subscribe

Artificial Intelligence

Google Opens Gemini 4 Carbon to Paying Developers and AI Ultra Subscribers

Google says what it calls its strongest model costs $2 per million input tokens and $10 per million output tokens through Nov. 30, then $4 and $20.

DailyMech Staff

Published

Last updated / Fact-checked on

Google Opens Gemini 4 Carbon to Paying Developers and AI Ultra Subscribers
Schäferle / Pixabay — Pixabay Content License

A dark, empty server-room aisle with glass-fronted cabinets and rows of blue indicator lights

Illustrative photo of a server-room aisle with glass-fronted cabinets and blue indicator lights; it is not Google, Gemini, or any data center named here. No logos or readable text are visible.

Google said Tuesday it is making Gemini 4 Carbon — which Google calls "our strongest model" — available through the Gemini API in Google AI Studio and Gemini for business for paying developers and enterprises. The introductory prices are $2 per million input tokens and $10 per million output tokens through Nov. 30, rising to $4 and $20 starting Dec. 1, according to Google's pricing page. The output price includes thinking tokens, and the model is not available on the free tier. It is also rolling out to Google AI Ultra subscribers in Google Antigravity, the Gemini app and AI Mode in Search.

Tuesday's release is separate from an earlier, limited one: Google said it released Carbon "two weeks ago" to a set of cybersecurity defenders via its Fairwind Program, which VentureBeat dated to Oct. 28. Carbon for cyber still requires applying to Fairwind. Google is also previewing a "/swarm" multi-agent feature in Antigravity.

At a glance

Item Detail Source/status
Availability Paid Gemini API in AI Studio and Gemini for business; Ultra subscribers in Antigravity, Gemini app, AI Mode in Search Google blog, Nov. 10, 2026 (fully read)
Intro pricing $2 input / $10 output per 1M tokens through Nov. 30; then $4 / $20 Google AI Studio pricing page (fully read)
Free tier Not available Google AI Studio pricing page (fully read)
Fairwind Defender release two weeks ago; cyber access still by application Google blog (fully read)
/swarm Preview multi-agent feature in Antigravity Google blog (fully read)
Context Up to 1M combined input-output tokens; March 2026 cutoff DeepMind model card (fully read)
Benchmark claim Google says #1 on CursorBench 4.0, WANDR, AutomationBench at >2X lower cost Google blog (fully read)

What Google released and where

Google said it is making Carbon available to developers, enterprises, scientists, and consumers, under the model code gemini-4-carbon, at "the same introductory price" as Gemini 4 Argon. Google said Argon "remains the ideal choice for the majority of software engineering and knowledge work," while Carbon uses more tokens than Argon.

According to the DeepMind model card, Carbon is based on Gemini 4 Argon and has a combined input-output token context window of up to 1M. It accepts text, images, audio, and video and outputs text, with a March 2026 knowledge cutoff. The card's cybersecurity results "reflect performance when accessed through the Fairwind Program."

VentureBeat reported that the rollout materials leave Vertex AI availability, regions, and support commitments unspecified, and that Google has yet to publish a country-by-country rollout schedule. VentureBeat described the release as making Carbon "generally available"; DailyMech did not find that phrasing in Google's post.

The benchmark claims, checked against Cursor's table

Google claims Carbon "ranks #1 on CursorBench 4.0, Perplexity's WANDR, and Zapier's AutomationBench at more than 2X lower cost per task compared to the next best external model." Cursor's public CursorBench 4.0 leaderboard lists Gemini 4 Carbon High first at 60.4%, Carbon Medium at 60.1%, and Anthropic's Fable 5.2 Max third at 60.0%. Cursor's page states: "Results are subject to variance; small differences in scores may not be statistically meaningful." DailyMech did not find a separate Cursor statement calling Carbon the top model; the ranking appears on the leaderboard.

VentureBeat reported earlier Tuesday that the CursorBench version it reviewed had no Carbon row and called Google's claim unverified. VentureBeat also reported that on Zapier's AutomationBench, Carbon high scored 53.88% versus Fable 5.2 max at 51.45%, and that Fable 5.2 leads Artificial Analysis's intelligence index at 65 to Carbon's 60. DailyMech did not independently verify the WANDR or AutomationBench tables.

How the price compares

VentureBeat lists Claude Fable 5.2 at $8/$40 and Claude Opus 5.5 at $4/$20 per million tokens, and said Carbon's intro rates are one-quarter of Fable 5.2's standard rates through Nov. 30 and half afterward. VentureBeat cautioned that "on some tasks, the model with the lower token price can cost more to run." DailyMech did not open Anthropic's pricing pages; those figures are VentureBeat's.

What to watch

Whether Cursor's leaderboard holds, given its variance caveat; whether WANDR and AutomationBench tables bear out Google's claims; any Vertex AI or regional details; and how Carbon's higher token use affects real costs compared with Argon.

How we reported this

DailyMech read Google's announcement on The Keyword, Google AI Studio's pricing page, and the DeepMind model card in full, along with Cursor's CursorBench page, Cursor's Carbon docs page, and VentureBeat's report by Carl Franzen. A Google pricing page at ai.google.dev timed out and was not used. DailyMech did not test the model and contacted no one.

Published and fact-checked November 10, 2026.

Sources

DailyMech used AI tools to help research and draft this article; it was fact-checked by our editors.

In Artificial Intelligence

From other sections

Sections

The Daily BriefingSample

Intelligence, distilled. Every morning.

Sample briefing. A five-minute read of the most important stories in AI and technology, once this list is real.

Sample form. It does not work. The address is not stored or sent.