🔍 Read the full analysis: One-Sixth The Cost, Second On MKA: Meet Gemini 4 Carbon on ThorstenMeyerAI.com
Get the little things that make your day 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 made Gemini 4 Carbon available to paying developers, businesses and Google AI Ultra subscribers on November 10, 2026, and published its model card. Carbon ranked second on the new MKA agent benchmark, narrowly behind Claude Fable 5.2, while Google DeepMind reported a substantially lower cost per MKA task; the benchmark was co-developed by Google and its tasks are private.
Google opened Gemini 4 Carbon to paying developers, businesses and Google AI Ultra subscribers on November 10, the same day Google DeepMind and Artificial Analysis introduced the MKA benchmark for AI agents working with visual and document-based material. Carbon placed second at 54.3%, behind Anthropic’s Claude Fable 5.2 at 56.6%, while Google DeepMind reported a much lower cost per task for Carbon.
The MKA benchmark tests agents on 200 expert-built tasks across 10 sectors, using material such as blueprints, scans and charts rather than plain text alone. Claude Fable 5.2 scored 56.6%, Gemini 4 Carbon 54.3% and Claude Opus 5.5 53.0%. The source report says these three results are within the margin of error, so the ranking does not establish a clear performance gap among them.
Google DeepMind’s report put Carbon’s cost at $5.63 per MKA task, compared with $35.28 for Fable 5.2. Those figures are specific to the benchmark and its reported evaluation; they do not by themselves establish what a company would pay for its own workloads. Google and Artificial Analysis launched MKA together, and the benchmark tasks are private, limiting outside verification of the results.
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 lists introductory API prices of $2 per million input tokens and $10 per million output tokens through November 30, 2026; prices rise to $4 and $20, respectively, on December 1.
Gemini 4 Carbon cuts benchmark task cost
Google opened Carbon to paying users on November 10, 2026. It placed second on the new MKA agent benchmark, with a reported task cost far below the leader’s.
A close benchmark result
The top three scores are reported within the margin of error, so the ranking does not establish a clear performance gap.
Claude Fable 5.2
Scored 56.6% on MKA.
Gemini 4 Carbon
Scored 54.3%, narrowly behind Fable.
Claude Opus 5.5
Scored 53.0% on MKA.
What the cost comparison shows
Google DeepMind reported benchmark-specific costs; actual business costs depend on each company’s workloads.
| Model | MKA task cost | MKA score |
|---|---|---|
| Gemini 4 Carbon | $5.63 | 54.3% |
| Claude Fable 5.2 | $35.28 | 56.6% |
Access and API pricing
Availability varies by product, and introductory API rates change after November.
Carbon’s Price and Performance Trade-Off
The launch gives companies another high-end model for work that may involve complex coding, extended reasoning and visual documents. Its second-place MKA result suggests it can handle the benchmark’s agent tasks competitively, but the narrow score difference and stated margin of error make a simple winner-takes-all reading inappropriate.
The reported MKA cost difference could matter to organizations running many such tasks, but it needs to be weighed against performance, speed and actual usage costs. Artificial Analysis describes Carbon as slow per task and verbose on its separate Intelligence Index evaluation: it reports 25.6 minutes per index task, compared with a 12.6-minute median for comparable models. That timing is not an MKA result, but it is a relevant trade-off for buyers considering the model.
Availability also depends on the product. Developers and businesses can access Carbon through paid Google services, while consumer access through the Gemini app and Search AI Mode requires Google AI Ultra. The launch is therefore not equivalent to unrestricted or free public access.
Products Worth Considering
Garmin Forerunner 165 GPS Running Smartwatch, 43 mm, Whitestone
Withings Sleep Under-Mattress Sleep Tracking Pad
Oral-B iO Series 3 Rechargeable Electric Toothbrush for Adults, Black, 3 Modes, 2 Brush Heads and Travel Case
As an affiliate, we earn on qualifying purchases.
A Specialist Model With a New Benchmark
Google describes Carbon as its top-tier Gemini 4 model, built on Gemini 4 Argon and intended for tasks that test a model’s reasoning capacity. Its model card says Carbon uses more thinking tokens than Argon to explore possibilities and verify work. Google also says Argon remains the better fit for most software engineering and knowledge-work tasks, positioning Carbon as a specialist rather than a universal replacement.
The model accepts text, images, audio and video, and produces text. Its context window can reach one million tokens across input and output, and Google lists a March 2026 knowledge cutoff. The company’s model card warns of hallucinations, occasional slowness or timeouts, and increased token use, particularly at higher thinking settings.
Google’s own model card reports Carbon leading on 15 of 17 benchmarks selected by Google, with GPT-6 Astra ahead on FrontierSWE 2 and Claude Opus 5.5 ahead on OSWorld-2.1. That scorecard is not the same as the newer independent comparison: Claude Fable 5.2, which the source says was released November 9, is absent from Google’s comparison table. Artificial Analysis’ current Intelligence Index also places Carbon below Fable 5.2, at 60 versus 65 at Fable’s maximum setting and 62 at its high setting.
““tasks that test the limits of a model’s intelligence””
— Google, in the Gemini 4 Carbon model card
Limits of the MKA Comparison
The MKA results are difficult for outsiders to reproduce because the tasks remain private. Google DeepMind co-developed the benchmark, and Carbon’s second-place result comes from a test in which a Google model is included. That does not invalidate the published scores, but it leaves independent researchers with limited means to inspect the task set or confirm the comparison.
The three leading MKA scores are reported to be within the margin of error; the source does not provide enough detail here to establish how often each model would win on repeated or varied tasks. It is also unclear whether the reported task costs include factors beyond the evaluation’s own accounting, or how those figures translate to real-world workloads.
Google’s 15-of-17 result comes from a company-selected set of benchmarks and rivals, while Artificial Analysis’ index uses a different group of tests. The scores should not be treated as directly interchangeable. Carbon’s launch-day availability also does not clarify whether access, pricing or performance will change after the introductory API period.
Pricing Changes on December 1
Developers considering Carbon can access it now through the paid Gemini API and Google AI Studio, with business access through Gemini for business. The published introductory API rates run through November 30, 2026. On December 1, Google’s listed rates increase from $2 to $4 per million input tokens and from $10 to $20 per million output tokens.
Further scrutiny of MKA will depend on what the benchmark’s organizers publish about its methods and results, since its tasks are not public. For buyers, the next practical test is whether Carbon’s capabilities, actual task costs and response times suit their own applications; the benchmark scores alone cannot answer that question.
Key Questions
Who can use Gemini 4 Carbon?
Google lists Carbon for paying users of the Gemini API, Google AI Studio and Gemini for business. Access in the Gemini app and AI Mode in Search is limited to Google AI Ultra subscribers.
How did Carbon perform on MKA?
Carbon scored 54.3%, placing second behind Claude Fable 5.2 at 56.6%. Claude Opus 5.5 scored 53.0%. The source report says the top three 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. These are benchmark-specific figures, not a guarantee of costs for other workloads.
What are Carbon’s listed API prices?
Through November 30, 2026, Google lists $2 per million input tokens and $10 per million output tokens. From December 1, the listed rates are $4 and $20, respectively.
Can outside researchers verify the MKA results?
Not fully from the information provided: MKA’s tasks are private, and Google DeepMind co-developed the benchmark. The published scores can be reported, but outsiders cannot inspect the task set themselves.
Source: ThorstenMeyerAI.com
