🔍 Read the full analysis: Gemini 4 Carbon Ranks Near The Top On MKA For Less on ThorstenMeyerAI.com
Get smart everyday buys 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 AI Ultra subscribers on November 10, while publishing its model card. On the new MKA benchmark, Carbon scored 54.3%, behind Claude Fable 5.2 at 56.6%; Google DeepMind’s report puts Carbon’s cost at $5.63 per task versus $35.28 for Fable, though the benchmark was co-developed by Google and its tasks are private.
Access depends on the product. Google made Carbon available through its 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 model card describes Carbon as a high-capability option for demanding work, including multi-step coding and scientific research.
Google’s published prices are $2 per million input tokens and $10 per million output tokens through November 30, 2026. From December 1, the listed rates rise to $4 and $20 per million tokens, respectively. These are token prices, not a fixed charge for each user task; the cost of a particular use depends on how much the model processes and produces.
On MKA, the reported scores were 56.6% for Claude Fable 5.2, 54.3% for Carbon and 53.0% for Claude Opus 5.5. The report says the top three results fall within the benchmark’s margin of error, so the narrow score gaps should not be read as a decisive separation. Google DeepMind reported a per-task cost of $5.63 for Carbon and $35.28 for Fable 5.2 on this test.
The MKA launch came alongside Google’s model card, which reports that Carbon led on 15 of 17 benchmarks selected by Google. Those results are not a universal ranking: Google chose the tests and competitors, the figures are dated to October 2026, and the comparison does not include Claude Fable 5.2, released November 9. On two listed tests, rivals led: GPT-6 Astra on FrontierSWE 2 and Claude Opus 5.5 on OSWorld-2.1.
Gemini 4 Carbon ranks near the top for less on MKA
Google’s model scored close to the leader on a new document-focused benchmark, with a lower reported task cost. The test’s private tasks and Google’s role in developing it limit outside scrutiny.
MKA results are close
The report says the top three scores fall within the benchmark’s margin of error.
What the cost comparison means
The per-task figures apply to MKA’s test tasks, not every customer workflow.
- Carbon’s reported MKA task cost was lower than Fable 5.2’s.
- That difference may matter for workflows with many runs.
- Carbon was described as slow and very verbose.
- Task mix, settings, response time and token use affect real costs.
Access depends on the product
The November 10 release opened paid access, with different routes for work and consumer use.
Available through the paid Gemini API, Google AI Studio and Gemini for business.
Gemini app and AI Mode in Search access is limited to Google AI Ultra subscribers.
Pricing is scheduled to rise
Google’s published rates are token prices; the cost of a task depends on how much the model processes and produces.
| Period | Input per million tokens | Output per million tokens |
|---|---|---|
| Through November 30, 2026 | $2 | $10 |
| From December 1, 2026 | $4 | $20 |
The Cost of Complex AI Tasks
The release gives businesses a new option for high-effort AI work, but the benchmark results point to trade-offs rather than a clear overall winner. Carbon’s MKA score was close to Fable 5.2’s, while the reported per-task cost was substantially lower. For organizations comparing models, that cost difference may matter when a workflow requires many runs, although the figure applies to the benchmark’s tasks and should not be treated as a universal price.
Carbon’s speed and token use also affect the practical comparison. Artificial Analysis describes it as slow per task and very verbose, reporting 25.6 minutes per index task against a 12.6-minute median for comparable models. Google’s model card warns of higher token use at higher thinking levels and occasional slowness or timeouts. A lower benchmark bill may therefore come with longer waits or more generated tokens, depending on the task and configuration.
For everyday users, access is more restricted than the broad developer launch might suggest: Gemini app and Search access requires AI Ultra. The announcement is most directly relevant to developers and businesses evaluating API performance, cost and reliability—not evidence that Carbon is the best choice for every user or workload.
Products Worth Considering
Logitech MX Master 4 Ergonomic Wireless Mouse with Haptic Feedback, Graphite
Brother INKvestment 6960 Wireless Color Inkjet All-in-One Printer with 11 x 17 Capabilities and 3.5 Color Touchscreen
Canon PIXMA PRO-200S Wireless Professional Inkjet Photo Printer
As an affiliate, we earn on qualifying purchases.
From Restricted Preview to Paid Access
Carbon was first made available to a small group of cyber defenders through Google’s Fairwind Program roughly two weeks before the November 10 paid release. Artificial Analysis lists October 28 as its release date. Google’s security-related capabilities remain limited to verified defenders, according to the source report.
Google describes Carbon as a model built on Gemini 4 Argon that uses more thinking tokens to work through hypotheses and extended verification. It accepts text, images, audio and video, and produces text. Its context window is up to one million tokens combined across input and output; its knowledge cutoff is March 2026, with Google warning that information may be less current in some domains. Google says Argon remains the better fit for most software engineering and knowledge work.
The independent Artificial Analysis Intelligence Index offers another comparison, although it is a separate measure from MKA. In version 5.0.1, Carbon at high thinking scored 60; Fable 5.2 scored 62 at high and 65 at its maximum setting. The index combines nine tests, so its ranking does not directly establish how a model will perform on a particular company’s documents or workflow.
““tasks that test the limits of a model’s intelligence””
— Google, in the Gemini 4 Carbon model card
Benchmark Limits and Open Questions
MKA’s results need independent scrutiny. Google DeepMind helped build the benchmark, and the tasks are private, so outside researchers cannot inspect the full test set or reproduce its results from the task materials. The published report gives scores and costs, but the source material does not provide enough detail to determine how closely the benchmark’s tasks match typical commercial workloads.
The close scores also come with a stated margin of error; the available figures do not establish that Fable 5.2 is reliably better than Carbon. The source material does not specify the margin’s numerical size. It is also unclear how the reported cost figures would change with different task mixes, model settings or usage patterns.
Google’s 15-of-17 scorecard has a separate limitation: Google selected the tests and rivals, and its results are dated to October. It excludes Fable 5.2, which was released November 9. Neither that scorecard nor MKA settles which model performs best across all uses. Carbon’s real-world reliability, speed and cost for individual customers remain to be measured against their own workloads.
Pricing Changes and Independent Testing
The next published pricing milestone is December 1, 2026, when Google’s listed rates are scheduled to move from $2/$10 to $4/$20 per million input/output tokens. Developers considering Carbon can compare it with Argon and other models using their own tasks, while accounting for access requirements, response times and token use rather than relying only on benchmark scores.
Further independent testing could clarify how well MKA predicts performance on real documents and whether the reported cost gap holds across different tasks. For now, the available evidence establishes that Carbon is commercially accessible through specified paid products, ranks near the top on the newly launched test, and has a lower reported MKA task cost than Fable 5.2. It does not establish a universal performance advantage or a guaranteed cost for customer deployments.
Key Questions
Who can use Gemini 4 Carbon now?
Google says 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.
How did Carbon perform on MKA?
Carbon scored 54.3%, placing second in the reported results behind Claude Fable 5.2 at 56.6% and ahead of Claude Opus 5.5 at 53.0%. The report says the top three are within the margin of error.
Is Carbon cheaper than Claude Fable 5.2?
On MKA, Google DeepMind’s report lists Carbon at $5.63 per task and Fable 5.2 at $35.28. Those are benchmark-specific figures, not a guaranteed price for every customer task. Token pricing also changes on December 1.
What are the main caveats about the MKA result?
Google co-developed the benchmark, its task set is private, and the source does not give the numerical margin of error. Those limits make it harder for outside groups to verify the test or determine how representative it is of real-world work.
Does Google’s scorecard show Carbon is the best model overall?
No. Google reports that Carbon led on 15 of 17 tests it selected, but rivals topped the other two, and the comparison did not include Claude Fable 5.2. The independent Artificial Analysis index placed Carbon behind Fable 5.2 in its current version.
Source: ThorstenMeyerAI.com
