🔍 Read the full analysis: Gemini 4 Carbon: Comparing Its MKA Position With The Leader’s Price 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 developers, businesses and 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 Fable’s $35.28. MKA was co-developed by Google DeepMind and Artificial Analysis, and its private tasks limit outside verification.
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 a benchmark for AI agents working with visual documents. Carbon placed second on the new MKA test, scoring 54.3% against Claude Fable 5.2’s 56.6%, while Google DeepMind reported a far lower cost per task for Carbon.
MKA tests whether AI agents can interpret blueprints, scans and charts, rather than relying only on plain text. Its 200 expert-built tasks span 10 sectors. 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 scores are within the benchmark’s margin of error, so the ranking does not establish a clear performance separation among them.
According to Google DeepMind’s report, Carbon’s average cost was $5.63 per MKA task, compared with $35.28 for Fable 5.2. These figures are tied to the benchmark’s reported task costs; they do not establish that Carbon will be cheaper across other workloads. The benchmark’s authors also include Google DeepMind, and its tasks are private, limiting independent inspection of how the test was constructed and scored.
Google describes Carbon as a higher-compute model for difficult, multi-step work, including complex coding and scientific research. Its model card reports top scores on 15 of 17 selected benchmarks, but those tests and comparison models were chosen by Google. The card’s results are dated October 2026 and do not include Fable 5.2, which the source says was released November 9. On Artificial Analysis’s independent Intelligence Index v5.0.1, Carbon scores 60, below Fable 5.2 at 62 on the high setting and 65 at maximum.
Gemini 4 Carbon Score, Cost and Trade-offs
Carbon scored close to the leader on a new visual-document benchmark, with a lower reported cost per task. The result is specific to MKA and its private tasks limit independent verification.
What MKA measures
The benchmark tests agents on visual documents such as blueprints, scans and charts.
Expert-built tasks
MKA includes 200 tasks across 10 sectors, developed by Google DeepMind and Artificial Analysis.
Close top scores
The report says the top three scores fall within the margin of error, so the ranking does not show a clear performance gap.
Broader capability and speed
Independent index results put Carbon behind Fable 5.2, while Carbon’s measured task time was longer than the comparable-model median.
| Measure | Gemini 4 Carbon | Claude Fable 5.2 |
|---|---|---|
| Intelligence Index score | 60 | 62 at high setting; 65 at maximum |
| Index cost per task | $7.76 | $9.08 at high setting |
| Task time | 25.6 minutes | Comparable-model median: 12.6 minutes |
Access and pricing
Availability depends on Google’s paid products, with introductory API rates scheduled to change.
Cost and Capability Trade-offs
The results frame Carbon as a possible lower-cost option for visually grounded agent tasks, not the top performer on MKA. For businesses evaluating agents that must extract information from diagrams or scanned documents, the reported price difference could affect how many tasks they can run within a budget. But the benchmark score and cost are specific to MKA; they should not be treated as a general forecast for every company’s workload.
Carbon’s broader standing is also mixed. It leads Google’s chosen model-card comparisons on many listed tests, while the newer independent index places it behind Fable 5.2. Artificial Analysis reports that Carbon costs $7.76 per index task at its standard price, less than Fable 5.2 at the high setting ($9.08) and Claude Opus 5.5 ($10.04), but more than GPT-6 Astra ($6.24) and GPT-6.1 Sol ($1.27). Its index results also come with a speed trade-off: the source says Artificial Analysis measured 25.6 minutes per task, against a 12.6-minute median for comparable models.
For buyers, the decision is not simply which model has the highest score. Price, response time, task type and access conditions all matter. Google says its less expensive sibling, Gemini 4 Argon, remains suited to most software engineering and knowledge work, positioning Carbon for harder cases where longer reasoning may be worth the extra time and token use.
Products Worth Considering
DJI Mini 4K Drone with 4K UHD Camera
Logitech MX Ergo S Advanced Wireless Trackball Mouse USB-C – Graphite
Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite
As an affiliate, we earn on qualifying purchases.
Release, Access and Pricing
Before its broad paid release, Carbon was made available to a small group of cyber defenders through Google’s Fairwind Program, about two weeks earlier, according to the source report. Artificial Analysis lists October 28 as its release date. The source says Carbon’s stronger security features remain restricted to verified defenders; it does not detail which features those are or how access is assessed.
As of the November 10 launch, Carbon was available through the paid Gemini API, Google AI Studio and Gemini for business. In the Gemini app and AI Mode in Search, it was 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, 2026. From December 1, the listed rates rise to $4 per million input tokens and $20 per million output tokens.
Google’s model card says Carbon accepts text, images, audio and video, and generates text. It supports a context window of up to one million tokens combined across input and output, with low, medium and high thinking settings; medium is the default. Google lists March 2026 as its knowledge cutoff and warns that knowledge may be less current in some domains. Its stated limitations include hallucinations, occasional slowness or timeouts, and higher token use, especially at higher thinking settings.
““tasks that test the limits of a model’s intelligence””
— Google, in the Gemini 4 Carbon model card
Limits of the Benchmark Evidence
MKA’s private tasks mean outside researchers cannot directly inspect the questions or reproduce the published results from the task set. Because Google DeepMind helped develop the benchmark and Carbon is a Google model, readers also lack a fully independent basis for judging whether the tasks favor particular systems. The reported closeness of the leading scores further makes a simple first-versus-second interpretation uncertain.
The available material does not explain the full method behind the reported per-task costs, including how task completion and model settings were standardized across providers. Nor does it establish how MKA performance translates to real business use. The launch pricing changes on December 1, while the source does not clarify whether those prices will remain fixed afterward. Carbon’s speed and token use may also vary by prompt, task and thinking level.
Pricing and Independent Testing
The next immediate pricing milestone is December 1, 2026, when Google’s listed API rates are scheduled to double from the introductory prices. Developers and businesses can test Carbon through Google’s paid offerings, subject to the stated access limits, and compare its cost, speed and results on their own tasks.
Further independent evaluation will help establish whether the MKA ranking holds across other visual-document tasks and whether its reported cost gap carries over to production use. Until researchers can inspect or independently reproduce the benchmark, its score is best read alongside the other available evidence, including Artificial Analysis’s index and Google’s dated model-card comparisons.
Key Questions
What happened with Gemini 4 Carbon?
Google opened Gemini 4 Carbon to paying developers, businesses and AI Ultra subscribers on November 10, 2026, and published its model card.
Did Carbon lead the new MKA benchmark?
No. 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 margin of error.
How much did Carbon cost per MKA task?
Google DeepMind’s report put Carbon at $5.63 per task, versus $35.28 for Fable 5.2. The figures refer to MKA and do not establish costs on other workloads.
Can anyone use Gemini 4 Carbon?
Access is paid and varies by product. Carbon is available through the Gemini API, Google AI Studio and Gemini for business; in the Gemini app and AI Mode in Search, the source says it is limited to Google AI Ultra subscribers.
What makes the MKA results uncertain?
Google DeepMind co-developed MKA, and the benchmark’s tasks are private. That limits outside review and independent reproduction of the results, even as the published scores offer a comparison between the listed models.
Source: ThorstenMeyerAI.com
