🔍 Read the full analysis: Gemini 4 Carbon Is Available: How Its MKA Rank Stacks Up 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 paid developers, businesses and Google AI Ultra subscribers on Nov. 10, 2026. Carbon scored 54.3% on the new MKA benchmark, behind Claude Fable 5.2 at 56.6%; Google DeepMind reported Carbon’s cost per MKA task at $5.63, compared with $35.28 for Fable.
Google opened Gemini 4 Carbon to paid developers, businesses and Google AI Ultra subscribers on November 10, while Google DeepMind and Artificial Analysis published results for a new benchmark of AI agents working with visual documents. Carbon scored 54.3% on MKA, behind Anthropic’s Claude Fable 5.2 at 56.6%, but Google DeepMind reported a substantially lower cost per task for Carbon.
MKA, short for the benchmark named in the source report, tests agents on 200 expert-built tasks across 10 sectors. The tasks involve material such as blueprints, scans and charts rather than only plain text. Claude Fable 5.2 scored 56.6%, Gemini 4 Carbon 54.3% and Claude Opus 5.5 53.0%. The report says the three results are within the benchmark’s margin of error, so the scores do not establish a clear performance separation.
Google DeepMind’s report put Carbon’s cost at $5.63 per MKA task, compared with $35.28 for Fable 5.2. That cost comparison is tied to this benchmark and the report’s method; it should not be read as a general price for every task or deployment. The benchmark was built by Google DeepMind and Artificial Analysis, and Google’s participation is relevant when interpreting a result that places its model near the top.
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. The listed price is $2 per million input tokens and $10 per million output tokens through Nov. 30; the source says those rates rise to $4 and $20, respectively, on Dec. 1, 2026.
Gemini 4 Carbon Is Available How Its MKA Rank Stacks Up
Google opened Carbon to paid users on November 10, 2026. Its MKA result is close to the top, with a much lower reported benchmark cost than Claude Fable 5.2.
A close benchmark race
The three leading scores fall within MKA’s margin of error, so they do not show a clear performance separation.
| Model | MKA score |
|---|---|
| Claude Fable 5.2 | 56.6% |
| Gemini 4 Carbon | 54.3% |
| Claude Opus 5.5 | 53.0% |
What MKA tests
The benchmark assesses agents handling visual documents, not only plain text.
MKA covers 200 tasks across 10 sectors.
Tasks use materials such as blueprints, scans and charts.
Access and listed pricing
Paid developers and businesses can access Carbon through Google’s developer and business products.
- Gemini API and Google AI Studio
- Gemini for business
- Gemini app and Search AI Mode for Google AI Ultra subscribers
- Input: $4 per million tokens, up from $2
- Output: $20 per million tokens, up from $10
Carbon’s Benchmark Lead Comes at a Cost
The release gives developers and businesses a new option for complex, multi-step work, but the benchmark results suggest a trade-off rather than a clear across-the-board lead. Carbon ranks just behind Fable 5.2 on MKA, while the reported per-task cost is far lower. For organizations running many document-heavy agent tasks, cost may shape model choice alongside score, speed, reliability and the cost of checking the output.
Those figures are not a guarantee of savings in real deployments. The source also reports that Artificial Analysis described Carbon as slow per task and very verbose: its index task took 25.6 minutes, compared with a 12.6-minute median for comparable models. Actual expenses and completion times will depend on task design, settings and usage. Buyers will need to test their own workloads rather than assume the MKA cost gap will carry over directly.
The change in access also makes Carbon’s launch more consequential than a leaderboard update. A model previously offered to a small group of cyber defenders is now available to paying customers through several Google products. However, the stronger security features described in the source remain limited to verified defenders, and the broad release does not mean every feature is open to every user.
Products Worth Considering
Apple Watch Series 8 GPS 41mm Midnight Aluminum Case with Midnight Sport Band
Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite
Bose TV Speaker Compact Soundbar with Bluetooth and HDMI-ARC, Black
As an affiliate, we earn on qualifying purchases.
From Limited Preview to Paid Access
Google describes Carbon as its top-tier Gemini 4 model, built on Gemini 4 Argon for work that tests a model’s limits, including complex coding and open-ended scientific research. Its model card says Carbon uses more “thinking tokens” than Argon for extended reasoning and verification. Google also says Argon remains the better fit for most software engineering and knowledge work, framing Carbon as a specialist rather than a universal replacement.
According to Google’s model card and developer guide, Carbon accepts text, images, audio and video, produces text, and supports a context window of up to 1 million tokens for input and output combined. Its thinking settings are low, medium and high, with medium as the default. Google lists hallucinations, occasional slowdowns or timeouts, and higher token use—especially at higher thinking settings—as known limitations.
Carbon was first made available to a small group of cyber defenders through Google’s Fairwind Program about two weeks before the paid release; Artificial Analysis lists Oct. 28 as its release date. Google’s model card reports top results on 15 of 17 selected benchmarks, but those comparisons used tests and rival models selected by Google and were dated October. The card did not include Fable 5.2, which the source says was released Nov. 9.
Benchmark Limits and Open Questions
The MKA task set is private, according to the source report, so independent researchers cannot inspect the tasks and reproduce the scores directly. Google DeepMind helped build the benchmark, and the reported results place Google’s model close to the top. The source says the leading scores are within the margin of error, but it does not provide the margin’s numerical size in the supplied material.
It is also unclear how closely MKA’s cost-per-task figures reflect costs for real customer workloads. The source does not give enough detail here to establish that the same cost relationship would hold across different task types, model settings or product plans. Nor does the benchmark score alone establish comparative accuracy, reliability or completion time for a particular business use.
Google’s 15-of-17 result is based on tests it selected and comparisons dated October 2026; it excludes Fable 5.2. Artificial Analysis’ newer index places Carbon behind Fable 5.2, but that index combines a different set of tests. These results answer different questions and should not be treated as interchangeable rankings.
Pricing Changes and Independent Testing
For users considering Carbon, the next confirmed date is December 1, 2026, when the listed API rates rise from $2 per million input tokens and $10 per million output tokens to $4 and $20. Developers can access the model through Google’s paid developer products now; availability in the Gemini app and Search AI Mode remains tied to Google AI Ultra.
Further evidence will depend on evaluations beyond the published MKA results, including testing on workloads that users can inspect and reproduce. The source does not give a date for another benchmark release or broader access to Carbon’s restricted security features. For now, customers must weigh the published scores and price figures against Google’s stated limitations and the time Carbon takes on Artificial Analysis’ index task.
Key Questions
Who can use Gemini 4 Carbon now?
Paid developers and businesses can use it through the Gemini API, Google AI Studio and Gemini for business. In the Gemini app and AI Mode in Search, the source says 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 says the three scores fall within the benchmark’s 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. This is a benchmark-specific comparison, not a confirmed price relationship for every workload.
When does Carbon’s listed price change?
The source lists rates of $2 per million input tokens and $10 per million output tokens through November 30, 2026. From December 1, it says the rates become $4 and $20, respectively.
What is not yet established about MKA?
The task set is private, limiting outside reproduction of the results. The supplied source also does not establish whether the benchmark’s cost figures predict costs across real customer workloads.
Source: ThorstenMeyerAI.com
