🔍 Read the full analysis: Gemini 4 Carbon Is Available At One-Sixth The Leader’s Cost on ThorstenMeyerAI.com
Get movie nights delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Google opened Gemini 4 Carbon to paying developers, businesses and Google AI Ultra subscribers on Nov. 10, 2026. On a new benchmark co-developed by Google DeepMind and Artificial Analysis, Carbon ranked second behind Claude Fable 5.2, while Google reported a much lower cost per task. The benchmark’s tasks are private, and the companies said the top three results fall within the margin of error.
Google opened Gemini 4 Carbon to paying developers, businesses and Google AI Ultra subscribers on November 10, as a new benchmark placed the model second behind Anthropic’s Claude Fable 5.2 on tasks involving charts, scans and blueprints. Google DeepMind reported that Carbon completed those benchmark tasks for $5.63 per task, compared with $35.28 for Fable 5.2, though the benchmark was co-developed by Google and its tasks are not public.
Carbon is available through the paid Gemini API, Google AI Studio and Gemini for business. In the Gemini app and AI Mode in Search, access is limited to Google AI Ultra subscribers. Google set introductory API prices at $2 per million input tokens and $10 per million output tokens through November 30. From December 1, the listed rates rise to $4 and $20 per million tokens, respectively.
The new benchmark, called MKA, was launched by Google DeepMind and testing company Artificial Analysis. Its 200 expert-built tasks span 10 sectors and require models to interpret material such as blueprints, scans and charts, rather than relying only on plain text. MKA results were 56.6% for Claude Fable 5.2, 54.3% for Carbon and 53.0% for Claude Opus 5.5. The report says the top three scores are within the margin of error.
Google’s model card gives Carbon the leading result on 15 of 17 tests selected by Google, but that comparison does not include Fable 5.2, which was released later. On two tests, GPT-6 Astra led Carbon on FrontierSWE 2, while Claude Opus 5.5 led on OSWorld-2.1. Google says the model-card results are from October 2026.
Gemini 4 Carbon cuts task costs with close benchmark scores
Google opened Carbon to paid users on November 10, 2026. It ranked second on MKA, while Google reported a lower cost per task than the top-ranked model.
A close benchmark race
The report says the top three scores fall within the margin of error.
Who can use Carbon
Access depends on the product and subscription.
- Developers through the Gemini API and Google AI Studio
- Businesses through Gemini for business
- Gemini app access
- AI Mode in Search
API prices change December 1
Introductory rates apply through November 30; listed rates then double.
| Per million tokens | Through Nov. 30 | From Dec. 1 |
|---|---|---|
| Input | $2 | $4 |
| Output | $10 | $20 |
What the comparison leaves open
The benchmark helps frame a decision, but does not settle which model fits a business workload.
- MKA tasks are private, limiting independent reproduction.
- Task costs may vary with prompts, settings, input size and retries.
- Carbon is reported as slow per task and may use more tokens at higher thinking levels.
Lower Costs, Closely Matched Scores
The MKA results make cost per completed task a central part of the comparison. Google DeepMind’s report puts Carbon’s MKA cost at $5.63 per task and Fable 5.2’s at $35.28. If those estimates hold for relevant workloads, businesses evaluating AI agents may weigh the price difference alongside accuracy, speed and the difficulty of their own tasks. The scores alone do not establish that one model is better for every use.
Cost comparisons also depend on how a model is configured and what work it performs. Artificial Analysis lists Carbon at $7.76 per Intelligence Index task at its standard price, below Claude Opus 5.5 at $10.04 and Fable 5.2 at its high setting at $9.08, but above GPT-6 Astra at $6.24 and GPT-6.1 Sol at $1.27. These are different measurements from the MKA cost figures, so they should not be treated as interchangeable estimates.
Speed may matter to buyers as much as the bill. Artificial Analysis describes Carbon as slow per task and verbose, reporting 25.6 minutes per Intelligence Index task against a 12.6-minute median for comparable models. That could affect workloads where time to completion matters, even if a model’s price per task is lower.
Popular right now
Philips 3300 Series Fully Automatic Espresso Machine with Milk Frother, Grinder, Hot and Iced Coffee Presets, SilentBrew, 15-Bar Pressure (E
Shark ZU102 Rotator Pet Upright Vacuum with PowerFins HairPro & Odor Neutralizer Technology, Charcoal
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Bluetooth 5.3, USB-C Charging, IPX4, Black
As an affiliate, we earn on qualifying purchases.
A Specialist Above Gemini 4 Argon
Google describes Carbon as its top-tier Gemini 4 model, built on Gemini 4 Argon for tasks that test a model’s capabilities, including complex multi-step coding and open-ended scientific research. According to Google’s model card, Carbon uses more thinking tokens than Argon to explore possibilities, reason and run verification steps. Google says Argon remains the better choice for most software engineering and knowledge-work tasks.
Carbon accepts text, images, audio and video, and produces text. Its context window can reach one million tokens across input and output, and users can select low, medium or high thinking levels, with medium as the default. Google lists a March 2026 knowledge cutoff, while warning that knowledge may be less current in some fields.
The model had earlier been made available to a small group of cyber defenders through Google’s Fairwind Program. Artificial Analysis lists October 28 as its release date. Google says Carbon’s stronger security features remain restricted to verified defenders. The November 10 opening expands access but does not make every feature or consumer-facing route available to all users.
“Argon remains the ideal choice for the majority of software engineering and knowledge work.”
— Google, in the Gemini 4 Carbon model card
Benchmark Limits and Access
MKA’s tasks are private, so outside researchers cannot inspect the test items or independently reproduce the results from the task set. Google DeepMind helped create the benchmark, and Google’s model performed strongly on it. That does not by itself invalidate the scores, but it is relevant when weighing the findings. The published report says the top three models are within the margin of error, so the small score gaps should not be read as a definitive ordering of capability.
It is also unclear how closely MKA’s task costs reflect a company’s own use. Costs can vary with prompts, model settings, input size and the number of attempts needed. Google lists hallucinations, occasional slowness or timeouts, and increased token use—especially at higher thinking levels—as known limitations. The supplied material does not establish how often these issues occur across real-world workloads.
Access is conditional: API and business availability requires paid access, while use in the Gemini app and Search’s AI Mode requires Ultra. The source material does not specify whether those access terms will change or how widely Carbon’s restricted security features may later be offered.
Pricing Changes on December 1
The next stated pricing milestone is December 1, 2026, when Google’s listed API rates are due to move from $2 per million input tokens and $10 per million output tokens to $4 and $20. Developers considering Carbon can compare results and costs during the introductory period, while accounting for its higher token use and possible delays.
Further independent scrutiny of MKA would help clarify how well the scores predict performance on business tasks, but the benchmark tasks are currently private. For now, the published comparisons show Carbon near the top of the measured field, not a clear lead: Fable 5.2 scored higher on MKA, the three leading results were within the margin of error, and other benchmarks produced different winners.
Key Questions
Who can use Gemini 4 Carbon now?
Paying developers and businesses can access it through the Gemini API, Google AI Studio and Gemini for business. Use in the Gemini app and AI Mode in Search is limited to Google AI Ultra subscribers.
How much does Gemini 4 Carbon cost?
Through November 30, Google lists API pricing at $2 per million input tokens and $10 per million output tokens. From December 1, the listed rates are $4 and $20 per million tokens.
Did Carbon win the new MKA benchmark?
No. Claude Fable 5.2 ranked first at 56.6%, Carbon scored 54.3%, and Claude Opus 5.5 scored 53.0%. The report says the top three results are within the margin of error.
Is Carbon cheaper than Claude Fable 5.2?
On MKA, Google DeepMind reported a cost of $5.63 per Carbon task, compared with $35.28 for Fable 5.2. That is a benchmark-specific estimate; it does not establish the cost difference for every task or use case.
What are Carbon’s known limitations?
Google lists hallucinations, occasional slowness or timeouts, and higher token use, particularly at higher thinking levels. Artificial Analysis also reported a 25.6-minute average per Intelligence Index task, compared with a 12.6-minute median for comparable models.
Source: ThorstenMeyerAI.com
