Google announced wider access to Gemini 4 Carbon on November 10, 2026, opening the model to paid developers and businesses and beginning a rollout to Google AI Ultra subscribers.
For businesses planning ongoing use, there is a near-term budget change: Standard Gemini Developer API rates are set to double on December 1. These are usage charges for connecting the model to software, not the price of an Ultra subscription.
Who can use Carbon
Carbon was first announced on October 28 with a restricted rollout to selected cybersecurity defenders and trusted partners. The new availability announcement opens access through the paid Gemini API in AI Studio and Gemini for business.
Google says Carbon is rolling out to Ultra subscribers in the Gemini app, AI Mode in Search and Antigravity, its coding platform. It also lists Ultra access in AI Studio. The plans page limits Carbon to selected countries; it does not promise Carbon to Pro or free subscribers. Check eligibility before upgrading specifically for the model.
Subscription access and API billing are different arrangements. Using a paid API key in AI Studio incurs usage charges, while Gemini app allowances depend on computational use, including prompt complexity and conversation length. Ultra access should not be read as an unlimited API entitlement.
What changes in December
Google’s Carbon-specific rate card lists the following Standard API prices in US dollars per million tokens, the units used to count model input and output. Carbon has no API free tier.
Billed usage | Through November 30, 2026 | From December 1, 2026 |
|---|---|---|
Uncached input | $2.00 | $4.00 |
Output, including thinking | $10.00 | $20.00 |
Cached input | $0.10 | $0.20 |
Thinking tokens are usage generated during the model’s internal reasoning, not just the answer you see. The cached-input discount applies to reused input that qualifies for implicit caching; it is not a discount on every request. Other API service tiers and enterprise contracts should be checked separately.
Illustrative calculation: one million uncached input tokens plus 100,000 billed output tokens, including thinking, cost $2 + $1 = $3 at the introductory rates. The same quantities cost $4 + $2 = $6 from December 1. This holds usage constant and excludes caching, tool charges, retries, discounts, taxes, infrastructure and human review.
The cost of a finished task still matters
Equal unit prices do not establish equal bills. Google’s model-selection guidance says Carbon and Gemini 4 Argon have the same per-token prices, but Carbon spends more tokens on internal reasoning and takes longer per turn. Google positions that extra work for demanding problems such as complex debugging and scientific reasoning.
For teams building integrations, there is another tradeoff: the output-token limit covers both thinking and the final answer. A tight cap can leave too little room for the response. Google offers low, medium and high thinking levels, with medium the default. Lowering that setting can reduce reasoning-token use, but Google warns it can also reduce quality on hard tasks.
Before committing a recurring workload, compare spending and elapsed time per acceptable result, including failed attempts, using December’s rates. Whether Carbon’s extra reasoning pays off depends on the task; the published prices alone cannot answer that question.
Readers evaluating models for repeated, routine work can also see RohitAI’s coverage of Muse Ember and task-level costs.