🔍 Read the full analysis: Gemini 4 Carbon Is Here: Second On MKA At A Lower Cost on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
Google opened Gemini 4 Carbon to paying developers, businesses and Google AI Ultra subscribers on November 10, 2026. It placed second on the new MKA benchmark, behind Anthropic’s Claude Fable 5.2, while Google DeepMind reported a lower cost per task; the benchmark was co-developed by Google and its tasks are not public.
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 the MKA benchmark for agents working with visual documents. Carbon ranked second at 54.3%, 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 5.2.
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. Google’s stated introductory prices are $2 per million input tokens and $10 per million output tokens through November 30. From December 1, those rates are scheduled to rise to $4 and $20, respectively.
MKA is designed to test AI agents on 200 expert-built tasks across 10 sectors, including work involving blueprints, scans and charts rather than only plain text. The published results put Fable 5.2 at 56.6%, Carbon at 54.3% and Claude Opus 5.5 at 53.0%. The source report says the scores for the top three are within the margin of error, so the small gaps should not be treated as definitive proof of a performance difference.
The cost figures come from Google DeepMind’s report: $5.63 for a Carbon task and $35.28 for a Fable 5.2 task on MKA. These are benchmark-specific costs, not a general estimate of what either model will cost for every real-world workload. Separately, Google’s model card reports Carbon leading on 15 of 17 selected benchmarks; it lists losses to GPT-6 Astra on FrontierSWE 2 and Claude Opus 5.5 on OSWorld-2.1.
Gemini 4 Carbon Is Here Second on MKA at a Lower Cost
Google opened Carbon to paying customers on November 10, 2026. It scored near the top on a visual document benchmark, with a lower reported task cost than the leader.
MKA leaderboard
The benchmark tests agents on visual, information-dense documents. The top three scores are within the margin of error.
| Model | MKA score | Reported task cost |
|---|---|---|
| Claude Fable 5.2 | 56.6% | $35.28 |
| Gemini 4 Carbon | 54.3% | $5.63 |
| Claude Opus 5.5 | 53.0% | Not stated |
Who can access Carbon
Access depends on the Google product and subscription.
- Gemini API and Google AI Studio
- Gemini for business
- Gemini app
- AI Mode in Search
Introductory API prices
Google’s listed rates are scheduled to double on December 1.
What the benchmark can show
MKA focuses on tasks involving materials such as blueprints, scans and charts.
Test your own workload
The results may inform evaluations of document-heavy agent work, but each organization needs to compare accuracy, speed and cost on its own tasks.
Tasks are private
Google co-developed MKA, and its tasks are not public, limiting independent inspection and reproduction.
Benchmark Rank Meets Lower Task Cost
The MKA result puts Carbon near the top of a test aimed at agents handling visual, information-dense material. That is relevant to businesses considering AI for tasks such as interpreting diagrams or extracting information from scans. But the result is one benchmark score, not evidence that Carbon will be the best choice across all business uses.
The reported cost gap could matter to organizations running repeated agent tasks. On MKA, Google DeepMind says Carbon’s task cost was far below Fable 5.2’s, even though Carbon scored slightly lower. Buyers will still need to compare accuracy, speed and cost on their own tasks. Artificial Analysis describes Carbon as slow per task and verbose: it took 25.6 minutes per Intelligence Index task, against a 12.6-minute median for comparable models.
Access and pricing also affect who can test the model. Developers and businesses can use it through paid Google products, while use in Google’s consumer app and Search AI Mode requires the Google AI Ultra subscription. The price increase scheduled for December makes the launch rate temporary rather than a long-term price guarantee.
Popular right now
Ninja Crispi Microwave & Air Fryer with FusionCrisp Technology, 14-in-1, Black (MW101BK1)
Shark NV752 Rotator Powered Lift-Away TruePet Upright Vacuum, Bordeaux
Ninja Crispi Microwave and Air Fryer with FusionCrisp, Silver, MW101
As an affiliate, we earn on qualifying purchases.
From Limited Release to Paid Access
Carbon is Google’s top-tier Gemini 4 model, built on Gemini 4 Argon. Google describes it as suited to demanding work such as complex, multi-step coding and open-ended scientific research. According to Google’s model card, Carbon uses more thinking tokens than Argon to explore hypotheses and run extended verification. Google also says Argon remains the better choice for most software engineering and knowledge work.
The model accepts text, images, audio and video and produces text. Its context window is up to one million tokens for input and output combined, with low, medium and high thinking settings. Google lists hallucinations, occasional slowness or timeouts, and higher token use—especially at higher thinking levels—as limitations. The model’s stated knowledge cutoff is March 2026, though Google cautions that information in some domains may be less current.
Carbon had previously been offered to a limited group of cyber defenders through Google’s Fairwind Program. The source material dates that restricted release to about two weeks before the November 10 wider paid launch. The model’s stronger security features remain limited to verified defenders, according to that material. Artificial Analysis’s current Intelligence Index v5.0.1 ranks Carbon (high) at 60; Fable 5.2 scores 62 at high and 65 at its maximum setting.
““tasks that test the limits of a model’s intelligence””
— Google DeepMind, in the Gemini 4 Carbon model card
Limits of the MKA Comparison
Google DeepMind co-developed MKA, and the benchmark tasks are private, according to the source material. That prevents outside researchers from independently inspecting the tasks or reproducing the results from the task set. The reported scores and costs therefore need to be read with those limits in mind.
The available material does not provide enough detail to establish how the margin of error was calculated or how closely MKA tasks reflect particular companies’ workloads. It also does not show whether the lower benchmark cost will translate into lower total costs in production, where task difficulty, repeated attempts, latency and human review can differ.
Google’s 15-of-17 model-card result is based on tests and competitors selected by Google, and the card’s comparison does not include Claude Fable 5.2, which the source says was released November 9. The model card reports results as of October 2026. The source material also does not establish whether Carbon’s speed or quality will change through future model updates.
Pricing Changes on December 1
The next clearly stated milestone is December 1, 2026, when Google’s listed API prices are scheduled to move from $2 per million input tokens and $10 per million output tokens to $4 and $20. Developers evaluating Carbon can compare performance and costs during the introductory period, but their results will depend on their prompts, workloads and chosen thinking settings.
Further clarity will depend on additional independent testing, more detail about benchmark methodology, and evidence from real deployments. Readers should also watch for updates to the model card and Artificial Analysis rankings. The current figures establish Carbon’s launch availability and published benchmark position; they do not settle how it compares across every task or what its operating costs will be for individual customers.
Key Questions
Who can use Gemini 4 Carbon?
It 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 rank on MKA?
Carbon scored 54.3%, second to Claude Fable 5.2 at 56.6%. Claude Opus 5.5 scored 53.0%. The source report says the top three results are within the margin of error.
Is Carbon cheaper than Claude Fable 5.2?
On MKA, Google DeepMind reported a cost of $5.63 per Carbon task and $35.28 per Fable 5.2 task. Those are benchmark-specific figures and do not establish costs for every type of use.
What will Gemini 4 Carbon cost?
Google’s stated rates through November 30 are $2 per million input tokens and $10 per million output tokens. From December 1, they are scheduled to rise to $4 and $20.
How independently verifiable is the MKA result?
Google DeepMind co-developed the benchmark, and the tasks are private, according to the source material. That limits outside review and replication of the published results.
Source: ThorstenMeyerAI.com
