🔍 Read the full analysis: A Closer Look At Gemini 4 Carbon’s MKA Performance And Price on ThorstenMeyerAI.com
Get the latest gadgets 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 November 10, 2026, and published its model card. On the new MKA benchmark, Carbon scored 54.3%, behind Claude Fable 5.2 at 56.6%; Google DeepMind reported Carbon cost $5.63 per task versus $35.28 for Fable, though the benchmark’s tasks are private and its creators include Google.
Google opened Gemini 4 Carbon to paying developers, businesses and Google AI Ultra subscribers on November 10, while publishing a model card and results from the new MKA benchmark. Carbon scored 54.3% on MKA, behind Anthropic’s Claude Fable 5.2 at 56.6%; Google DeepMind’s report put Carbon’s cost at $5.63 per task, compared with $35.28 for Fable.
MKA, developed by Google DeepMind and Artificial Analysis, tests AI agents on 200 expert-built tasks across 10 sectors. The tasks involve interpreting material such as blueprints, scans and charts, rather than only answering questions in plain text. Claude Fable 5.2 scored 56.6%, Gemini 4 Carbon 54.3%, and Claude Opus 5.5 53.0%. The source report says the top three results are within the benchmark’s margin of error, so the small gaps should not be read as definitive differences in capability.
Google’s model card reports Carbon leading on 15 of 17 selected benchmarks against four rivals. Those results are Google’s own comparison, based on tests and competitors it selected, and the card’s reported results are as of October 2026. It does not include Claude Fable 5.2, which was released November 9. Carbon trails GPT-6 Astra on FrontierSWE 2 and Claude Opus 5.5 on OSWorld-2.1.
Access depends on the product. Carbon is available through the paid Gemini API, Google AI Studio and Gemini for business; within the Gemini app and Search’s AI Mode, it is limited to Google AI Ultra subscribers. Google set API prices at $2 per million input tokens and $10 per million output tokens through November 30. From December 1, those rates rise to $4 and $20, respectively.
Gemini 4 Carbon’s MKA Performance and Price
Google opened Carbon to paying developers, businesses and Google AI Ultra subscribers on November 10, 2026. Its MKA score trails Claude Fable 5.2, while Google reports a lower cost per task.
What MKA measures
The benchmark tests agents on expert-built tasks involving visual and technical materials.
Beyond plain text
Tasks involve interpreting blueprints, scans and charts.
Many sectors
MKA includes 200 tasks across 10 sectors.
How the top scores compare
The report says the top three results fall within the benchmark’s margin of error.
| Model | MKA score |
|---|---|
| Claude Fable 5.2 | 56.6% |
| Gemini 4 Carbon | 54.3% |
| Claude Opus 5.5 | 53.0% |
Access depends on the product
- Developers: Gemini API and Google AI Studio
- Businesses: Gemini for business
- Gemini app: Google AI Ultra
- Search AI Mode: Google AI Ultra
API prices change December 1
Cost Versus Benchmark Position
The results make Carbon a potential option for organizations weighing performance against the cost of complex agent tasks. In Google DeepMind’s MKA report, Carbon’s per-task cost is substantially below Fable 5.2’s, even though Fable scored higher. That comparison may interest businesses testing systems that must interpret visual or technical documents, where a task’s total cost can matter as much as a score.
The figures need qualification. The MKA cost comparison comes from Google DeepMind’s report, and the benchmark was co-developed by Google and Artificial Analysis. MKA’s tasks are private, limiting outside scrutiny and independent replication. Its result is one measure of performance, not proof that Carbon will be cheaper or more accurate for every company’s workload.
Carbon’s price also changes soon after launch. The introductory API rates last only until November 30, and a higher price could alter the cost comparison for users who adopt it after December 1. The published per-task MKA figure and the API price schedule describe different ways of measuring expense, so they should not be treated as interchangeable.
Products Worth Considering
Samsung Galaxy Tab A9+ 11-Inch Android Tablet, 64GB, Graphite
Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite
LG UltraGear 32GX850A-B 32-inch 4K UHD OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
A Specialist Model, Now Paid
Google describes Carbon as its top-tier model built on Gemini 4 Argon, intended for demanding work such as multi-step coding and debugging or open-ended scientific research. According to its model card, Carbon uses more thinking tokens than Argon to explore possibilities and verify results. Google also says Argon remains the better choice for most software engineering and knowledge work, framing Carbon as a model for harder cases rather than a general replacement.
Carbon accepts text, images, audio and video, and produces text. Its context window is up to one million tokens, counting input and output together. Google lists low, medium and high thinking levels, with medium as the default. The company identifies hallucinations, occasional slowness or timeouts, and higher token use—especially at higher thinking levels—as known limitations.
Before the broader paid release, Carbon was made available to a small group of cyber defenders through Google’s Fairwind Program. The source report says that rollout began about two weeks earlier and that enhanced security features remain restricted to verified defenders. The November 10 release widened access, but did not make every feature available to every user.
““tasks that test the limits of a model’s intelligence””
— Google, in the Gemini 4 Carbon model card
Limits of the Published Comparisons
MKA’s private task set means outside researchers cannot inspect the questions or independently reproduce the results. Google DeepMind helped create the benchmark, and its report supplies the cost comparison. Those facts do not invalidate the scores, but they limit what readers can verify from the published information.
It is also unclear how closely MKA results predict performance on any particular company’s documents, software or operating procedures. The reported results give percentages and a per-task cost, but the source material does not specify enough detail to establish that every task has the same real-world value or expense. The stated margin of error further cautions against treating the small score differences among the top three as a settled ranking.
Google’s 15-of-17 model-card result is also not a complete independent comparison: Google selected the tests and rival models, and the table excludes Claude Fable 5.2. Artificial Analysis’s separate Intelligence Index puts Carbon at 60 in version 5.0.1, compared with Fable 5.2 at 65 at maximum settings and 62 at high. The source report also describes Carbon as slow on that index, at 25.6 minutes per task versus a 12.6-minute median for comparable models. These measures address different evaluations and should not be collapsed into a single overall ranking.
Access and Pricing After November
Developers and businesses can use Carbon through Google’s paid developer and business offerings, while Gemini app and Search users need Google AI Ultra. The next clear date is December 1, 2026, when the listed API rates double from the introductory $2 per million input tokens and $10 per million output tokens to $4 and $20.
Users evaluating the model can compare its output, speed and token consumption on their own workloads, rather than relying only on benchmark averages. Google’s own model card identifies slow responses and higher token use as limitations, while Artificial Analysis reports a longer-than-median time per index task. The available source material does not specify further releases or changes to MKA, so any broader follow-up on independent testing, public task access or Carbon’s availability remains unannounced.
Key Questions
Who can use Gemini 4 Carbon?
Paid developers and businesses can access it through the Gemini API, Google AI Studio and Gemini for business. In the Gemini app and Search’s AI Mode, access is limited to Google AI Ultra subscribers.
How did Carbon score on MKA?
Carbon scored 54.3%, behind Claude Fable 5.2 at 56.6% and ahead of Claude Opus 5.5 at 53.0%. The source report says the top three are within the benchmark’s margin of error.
How much does Carbon cost?
Through November 30, 2026, Google lists API rates of $2 per million input tokens and $10 per million output tokens. Starting December 1, the rates become $4 and $20. Google DeepMind separately reported a $5.63 cost per MKA task.
Can outside researchers verify the MKA results?
Not fully from the available information. MKA’s 200 tasks are private, and Google DeepMind helped develop the benchmark. That limits independent inspection and replication of the task set.
Does Google recommend Carbon for everyday work?
Google says Gemini 4 Argon remains its preferred option for most software engineering and knowledge work. It describes Carbon as intended for harder, more demanding tasks and lists slower responses and higher token use among its limitations.
Source: ThorstenMeyerAI.com
