🔍 Read the full analysis: Could Gemini 4 Carbon’s Price Change The MKA Equation? 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 Google AI Ultra subscribers on November 10, 2026. On the new MKA benchmark, it scored 54.3%, behind Claude Fable 5.2 at 56.6%, while Google DeepMind reported a much lower cost per task for Carbon; the benchmark was co-developed by Google and its tasks are private.
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 non-text materials. Carbon scored 54.3% on MKA, behind Anthropic’s Claude Fable 5.2 at 56.6%, but Google DeepMind reported a far lower per-task cost for Carbon: $5.63 versus $35.28. The comparison could matter to organizations weighing performance against expense, though the benchmark’s tasks are private and Google helped create it.
Access depends on the product. Carbon is available through the paid Gemini API, Google AI Studio and Gemini for business; in the Gemini app and AI Mode in Search, it is limited to Google AI Ultra subscribers. Google’s introductory API pricing is $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.
On MKA, which comprises 200 expert-built tasks across 10 sectors, Claude Fable 5.2 scored 56.6%, Carbon 54.3% and Claude Opus 5.5 53.0%. The source report describes the scores of the top three as being within the margin of error, so the ranking should not be read as a decisive separation. Google DeepMind’s report puts Carbon’s cost at $5.63 per task, compared with $35.28 for Fable 5.2. Those are reported benchmark costs, not a guarantee of what a particular customer will pay for real workloads.
Other evaluations give a mixed picture. Google’s model card says Carbon led on 15 of 17 benchmarks selected for its comparison, but those results were as of October 2026 and did not include Fable 5.2, released November 9. On Artificial Analysis’ independent Intelligence Index v5.0.1, Carbon at its high setting scored 60, behind Fable 5.2 at 62 on high and 65 at its maximum setting. Artificial Analysis also reported that Carbon took 25.6 minutes per index task, compared with a 12.6-minute median for comparable models.
Could Gemini 4 Carbon’s Price Change the MKA Equation?
Carbon scored close to the MKA leader while Google reported a much lower cost per task. For businesses, the trade-off depends on quality, speed and the cost of their own workloads.
Close scores, different costs
The reported score gap is small, and the source says the top three results fall within the margin of error.
| Model | MKA score | Cost per task |
|---|---|---|
| Claude Fable 5.2 | 56.6% | $35.28 |
| Gemini 4 Carbon | 54.3% | $5.63 |
| Claude Opus 5.5 | 53.0% | Not stated |
What the benchmark tests
MKA evaluates AI agents on complex visual materials rather than text alone.
The benchmark includes 200 expert-built tasks across 10 sectors.
Tasks use materials such as blueprints, scans and charts.
The task set is private, limiting outside scrutiny and independent reproduction.
Performance has other trade-offs
Independent evaluations and Google’s model card offer different views of Carbon’s strengths.
- Google’s model card says Carbon led on 15 of 17 selected benchmarks.
- Lower reported MKA task cost could matter for teams running many similar tasks.
- Artificial Analysis ranked Carbon behind Fable 5.2 on its Intelligence Index.
- Carbon took 25.6 minutes per index task, versus a 12.6-minute comparable-model median.
- Google lists higher token use, occasional slowness and timeouts as limitations.
Access and API pricing
The launch is available through paid developer and business products, with app access limited to Ultra subscribers.
Cost Versus Performance on MKA
The headline comparison is not simply which model scored highest. For teams running many document-heavy or visual tasks, the reported difference in cost per MKA task may affect whether a small score gap is worth paying for. If Carbon can handle a given workload at a substantially lower cost, some organizations may prefer it even when another model leads the benchmark. That is a possible procurement consideration, not proof that Carbon is the better choice across applications.
The scores and prices also need to be read together with practical limits. Carbon is designed to spend more time reasoning, and Google’s documentation lists higher token use, occasional slowness and timeouts among its limitations. Artificial Analysis’ reported task time adds another operational measure: a lower benchmark bill may not mean a faster result. Businesses will need to compare quality, latency and costs on tasks resembling their own work rather than assume the MKA figures will transfer directly.
Availability is another part of the equation. Carbon’s launch moves it beyond a limited early-access group, but app and Search access remains restricted to Ultra subscribers, and the promotional API rate is temporary. The announcement gives developers and companies a new model to test; it does not establish broad consumer access or show how much demand it will attract.
Products Worth Considering
LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1, DisplayPort 1.4, Black
Dell UltraSharp U2424H 24-Inch FHD Monitor
Samsung 27-inch Odyssey OLED G6 QHD Gaming Monitor
As an affiliate, we earn on qualifying purchases.
A New Model and Benchmark
Carbon is Google’s top-tier model built on Gemini 4 Argon. Google describes it as suited to demanding work such as multi-step coding and debugging and open-ended scientific research. According to Google’s model card, it uses more thinking tokens than Argon to examine possible answers and run extended checks. Google also says Argon remains the better fit for most software engineering and knowledge work, positioning Carbon as a specialist rather than a replacement for every task.
The model accepts text, images, audio and video, and produces text. Its context window is up to one million tokens, with input and output combined; its stated knowledge cutoff is March 2026. Google lists hallucinations, occasional slowness or timeouts, and higher token use as known limitations. Carbon first went to a small group of cyber defenders through Google’s Fairwind Program about two weeks before the wider paid release. Its stronger security features remain restricted to verified defenders, according to the source report.
MKA was launched by Google DeepMind and Artificial Analysis to test agents on blueprints, scans and charts rather than text alone. The test’s subject matter makes it relevant to work involving complex visual documents, but its private task set limits outside scrutiny. That matters when interpreting a result from a benchmark partly developed by the company whose model is being evaluated.
““tasks that test the limits of a model’s intelligence””
— Google, in the Gemini 4 Carbon model card
Benchmark Limits and Open Questions
The MKA results are not independently reproducible by outsiders because the tasks remain private. Google DeepMind helped build the benchmark, and Carbon is Google’s model; that does not establish that the results are biased, but it gives readers reason to seek validation from other tests. The source report says the leading scores are within the margin of error, but does not provide further detail here about the size of that margin or the statistical method.
It is also unclear how well MKA’s cost-per-task figures predict bills for customer workloads. Actual expenses can depend on task complexity, prompt length, model settings and how often a system needs to retry or check its work. The reported comparison does not settle whether Carbon’s lower benchmark cost offsets its slower processing for a particular organization.
Google’s own 15-of-17 benchmark tally reflects tests and rival models selected for its model card, while the independent index uses a different mix of evaluations and ranks Carbon behind Fable 5.2. These results are not directly interchangeable. Carbon’s real-world reliability, demand at the new access tiers and performance on independently designed multimodal tasks remain to be established.
Pricing Changes and Wider Testing
Developers and businesses can test Carbon through Google’s paid developer and business offerings, while Google AI Ultra subscribers can access it in the Gemini app and AI Mode in Search. The next concrete pricing milestone is December 1, 2026, when Google’s listed API rates are due to move from the introductory $2/$10 per-million-token prices to $4/$20.
Further comparisons will depend on independent evaluations and customer tests using representative tasks. Readers should watch for updated benchmark results, additional assessments of speed and cost, and evidence about how Carbon performs outside the private MKA task set. The available information does not specify a date for another model release or for public access to MKA’s tasks.
Key Questions
When did Gemini 4 Carbon become available?
Google opened it to paying developers, businesses and Google AI Ultra subscribers on November 10, 2026. Access differs by product: the Gemini app and AI Mode in Search require an Ultra subscription.
Did Carbon win the new MKA benchmark?
No. Carbon scored 54.3%, placing second behind Claude Fable 5.2 at 56.6%. The source report says the top three scores are within the margin of error.
How much does Carbon cost on MKA?
Google DeepMind’s report lists a $5.63 cost per MKA task for Carbon, compared with $35.28 for Fable 5.2. That benchmark figure does not establish the cost of a customer’s specific workload.
What are Carbon’s API prices?
Through November 30, Google lists $2 per million input tokens and $10 per million output tokens. From December 1, the rates are scheduled to rise to $4 and $20, respectively.
Can outsiders verify the MKA tasks?
Not from the information provided: MKA’s tasks remain private. Google DeepMind co-developed the benchmark, so independent assessments using other tests will be useful for judging how broadly its results apply.
Source: ThorstenMeyerAI.com
