🔍 Read the full analysis: What Gemini 4 Carbon’s Second-Place MKA Ranking Means on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Google opened Gemini 4 Carbon to paying users on November 10, the same day Google DeepMind and Artificial Analysis introduced the MKA benchmark for AI agents working with visual and document-based tasks. Carbon scored 54.3%, just behind Claude Fable 5.2 at 56.6%; the benchmark report says Carbon costs $5.63 per task versus $35.28 for Fable. The ranking is a useful comparison, but the top three results are within the stated margin of error, and Google helped build MKA while the test tasks remain private.
Google opened Gemini 4 Carbon to paying developers, businesses and Google AI Ultra subscribers on November 10, the day Google DeepMind and Artificial Analysis published results from their new MKA benchmark for AI agents working with blueprints, scans and charts. Carbon placed second at 54.3%, behind Anthropic’s Claude Fable 5.2 at 56.6%, while the report put Carbon’s cost at $5.63 per task compared with $35.28 for Fable.
MKA evaluates agents on 200 expert-built tasks across 10 sectors. The published results put Claude Fable 5.2 first at 56.6%, Gemini 4 Carbon next at 54.3%, and Claude Opus 5.5 third at 53.0%. The report says the three scores fall within the margin of error, so the small gaps do not establish a clear performance separation. Google DeepMind’s report lists a cost of $5.63 per Carbon task and $35.28 per Fable 5.2 task.
Carbon is Google’s top-tier model, built on Gemini 4 Argon and designed for difficult, multistep work such as coding and scientific research. Google’s model card says it uses more thinking tokens than Argon, which can mean higher token use and slower responses. Google also lists hallucinations and occasional slowdowns or timeouts among its limitations. The company says Argon remains the better choice for most software engineering and knowledge work.
Access is not identical across Google’s products. Carbon is available through the paid 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. The stated API price is $2 per million input tokens and $10 per million output tokens through November 30, 2026, rising to $4 and $20 from December 1. Google’s stronger security features remain restricted to verified defenders, according to the source report.
Gemini 4 Carbon’s MKA Ranking What second place means
Carbon scored close to the top on a new benchmark for visual and document tasks. Its reported lower cost may appeal to businesses, but the ranking does not establish a universal best choice.
A close ranking
The report says the top three scores fall within the margin of error.
Claude Fable 5.2
Scored 56.6%, a small lead that does not establish clear separation.
Gemini 4 Carbon
Scored 54.3% on tasks involving blueprints, scans and charts.
Claude Opus 5.5
Scored 53.0%, also within the stated margin of error.
What MKA measures
The benchmark tests AI agents on expert-built work across multiple sectors.
Tasks use blueprints, scans and charts, not only plain-text prompts.
The evaluation covers 200 tasks across 10 sectors.
Published scores and per-task estimates give buyers one point of comparison.
What the cost gap may mean
The reported benchmark costs favor Carbon, but real bills depend on each company’s workload.
- Lower reported cost per MKA task could matter for repeated analysis.
- Paying developers and businesses can test Carbon through Google’s developer and business channels.
- Compare accuracy, latency and total operating cost on your own tasks.
- Account for review or correction needs and Carbon’s reported slowdowns.
Limits buyers should keep in view
The published ranking has caveats that affect how far it can be generalized.
- Google DeepMind helped create MKA, and its test tasks are private.
- The close scores do not show which model will perform best in a particular workflow.
- Benchmark cost estimates do not guarantee the same gap on customer workloads.
- API prices are listed to rise on December 1, 2026.
What the Cost Gap Says
The MKA result gives buyers a comparison that pairs task performance with reported cost, rather than treating a model’s rank as the only measure. If the reported prices apply to a company’s own workload, Carbon’s lower per-task cost could make it attractive for repeated document and visual-analysis tasks, even without the highest score. But benchmark costs do not by themselves predict a customer’s bill: actual spending can depend on task design, usage volume, and how much review or correction a system needs.
The ranking also illustrates why a second-place result should be read with its uncertainty attached. Carbon’s score is close to Fable 5.2’s, and the report places the top three within the margin of error. For companies choosing an agent, MKA may provide one data point, but it does not settle which model is best for a particular workflow. Buyers would need to test their own tasks and compare accuracy, latency and total operating cost.
Popular right now
KitchenAid Artisan 5-Quart Tilt-Head Stand Mixer, Blue Velvet
De'Longhi Rivelia Automatic Espresso Machine with Grinder and Milk Frother, Beige
Ninja CREAMi 7-in-1 Ice Cream Maker, NC299AMZ
As an affiliate, we earn on qualifying purchases.
A New Test, a New Ranking
Carbon’s launch follows a short period of limited access. According to the source report, it first went to a small group of cyber defenders through Google’s Fairwind Program about two weeks before its broader paid availability. Artificial Analysis lists October 28 as its release date. The November 10 opening therefore moved the model from restricted access to paid developer and business use, while its security features remained limited to verified defenders.
The MKA result is separate from Google’s own model-card comparisons. Google reports Carbon leading on 15 of 17 benchmarks it selected, but that scorecard does not include Claude Fable 5.2, which was released November 9, and Google says the cited results are from October 2026. The outside Artificial Analysis Intelligence Index v5.0.1 gives Carbon a score of 60; Fable 5.2 scores 65 at its maximum setting and 62 at high. These are different evaluations, with different test designs, so their rankings should not be treated as interchangeable.
Artificial Analysis also describes Carbon as slow and verbose on its Intelligence Index: it took 25.6 minutes per index task, compared with a 12.6-minute median for comparable models. That is a separate measurement from MKA and does not establish how quickly Carbon would complete any specific business task.
Limits of the MKA Results
The MKA results come with a transparency limitation: Google DeepMind helped create the benchmark, and the source report says the test tasks are private, leaving outside researchers unable to inspect them. That does not invalidate the scores, but it limits independent checks of task selection and reproducibility. The report also says the top three are within the margin of error; it does not provide enough detail here to establish how often any model would win on a different set of tasks.
The cost figures are attributed to Google DeepMind’s report, and their meaning depends on the benchmark’s task and pricing assumptions. The available material does not establish whether the same cost gap would hold for customers’ real-world workloads. Nor does it show how Carbon compares with Fable 5.2 on every measure of speed, reliability or output quality. Those questions remain open beyond the published MKA score and per-task estimates.
Access, Pricing and Further Tests
Developers and businesses can now test Carbon through Google’s paid developer and business channels, subject to the product-specific access rules. They will need to account for the scheduled December 1 price increase when estimating ongoing use: the listed API rates double from $2 to $4 per million input tokens and from $10 to $20 per million output tokens. The source material does not identify a scheduled MKA retest or an independent release of its private tasks.
The next useful evidence will come from further published evaluations and from organizations testing Carbon against their own workloads. Those comparisons can show whether its lower reported MKA cost offsets its near-tied score and reported slowness in practical use. Until then, the result supports a limited conclusion: Carbon is competitive on this new benchmark, but the available data do not establish a universal best choice.
Key Questions
What did Gemini 4 Carbon score on MKA?
It scored 54.3%, placing second in the published results. Claude Fable 5.2 scored 56.6%, and Claude Opus 5.5 scored 53.0%; the report says the top three are within the margin of error.What does the MKA benchmark test?
MKA uses 200 expert-built tasks across 10 sectors to assess AI agents handling material such as blueprints, scans and charts, rather than only plain-text prompts.Is Gemini 4 Carbon cheaper than Claude Fable 5.2?
Google DeepMind’s report gives a benchmark cost of $5.63 per Carbon task, compared with $35.28 for Fable 5.2. Those are reported MKA costs and do not guarantee the same price difference for other workloads.Who can use Gemini 4 Carbon?
It is available to paying developers through the Gemini API and Google AI Studio, and to businesses through Gemini for business. In the Gemini app and AI Mode in Search, access is limited to Google AI Ultra subscribers.Does the MKA result prove Carbon is the best model for business?
No. The score places Carbon close to the top on one benchmark, but the three leading results are within the stated margin of error. Google helped build MKA, its tasks are private, and the published result does not show how each model performs on an individual company’s work.Source: ThorstenMeyerAI.com
