🔍 Read the full analysis: What You Get With Gemini 4 Carbon At One-Sixth The Leader’s 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, and published its model card. Carbon scored 54.3% on the new MKA benchmark, just behind Claude Fable 5.2 at 56.6%; Google DeepMind reported Carbon’s MKA task cost at $5.63 versus $35.28 for Fable 5.2. The benchmark was co-developed by Google DeepMind, and its private tasks limit independent verification.
Google opened Gemini 4 Carbon to paying developers, businesses and Google AI Ultra subscribers on November 10, while a new benchmark from Google DeepMind and Artificial Analysis ranked it second behind Anthropic’s Claude Fable 5.2. Carbon scored 54.3% on the benchmark, compared with Fable 5.2’s 56.6%; Google DeepMind’s report put Carbon’s cost at $5.63 per task, versus $35.28 for Fable. The result gives buyers a new price-performance comparison, though the benchmark’s private tasks and Google’s role in building it limit outside scrutiny, as discussed in the latest AI ranking analysis.
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 set introductory API prices at $2 per million input tokens and $10 per million output tokens through November 30. Starting December 1, those rates rise to $4 per million input tokens and $20 per million output tokens.
The MKA benchmark evaluates AI agents on 200 expert-built tasks across 10 sectors, using materials such as blueprints, scans and charts rather than only plain text. Fable 5.2 scored 56.6%, Carbon 54.3%, and Claude Opus 5.5 53.0%. The source report says the top three results are within the benchmark’s margin of error. Google DeepMind reported a per-task cost of $5.63 for Carbon and $35.28 for Fable 5.2 on MKA.
Google’s model card paints a different, broader picture: Carbon leads on 15 of 17 benchmarks Google selected. The card’s comparisons do not include Fable 5.2, which was released on November 9, and Google says the results are from October 2026. On Artificial Analysis’ Intelligence Index version 5.0.1, Carbon at its high setting scores 60; Fable 5.2 scores 62 at high and 65 at its maximum setting. These measures cover different tests and should not be treated as interchangeable rankings.
Gemini 4 Carbon at a lower reported task cost
Google opened Carbon to paying users on November 10, 2026. It scored just behind Claude Fable 5.2 on MKA, with a much lower reported cost per task.
Where Carbon is available
Access depends on the product and subscription.
- Gemini API
- Google AI Studio
- Gemini for business
- Gemini app
- AI Mode in Search
API price change
Introductory rates apply through November 30.
Built for demanding work
Google positions Carbon as a specialist model for complex tasks.
Multimodal input
Accepts text, images, audio and video, and produces text.
Large working window
Supports up to 1 million tokens across input and output.
More thinking, possible delays
Google notes higher token use, hallucinations and occasional slowness or timeouts.
What buyers should weigh
Benchmark results vary by test, and MKA has limits.
- MKA’s tasks are private, limiting independent verification.
- Google DeepMind co-developed the benchmark.
- The top three MKA results are within the reported margin of error.
- Test quality, latency and token use on your own workloads.
The Cost of Complex AI Tasks
The immediate appeal is the reported difference in cost per MKA task. If the benchmark’s tasks reflect work a buyer needs to do, Carbon could offer a lower-cost route to handling documents and visual materials that require an AI agent to interpret more than ordinary text. The result may matter to organizations comparing model quality against the cost of repeated or large-scale use.
That comparison is not a universal price guarantee. A per-task figure depends on the benchmark’s task design, model settings and how much work a model uses to reach an answer. Google describes Carbon as a model that spends more tokens thinking, and says it can use more tokens at higher thinking levels. Real-world costs and speed may differ by workload. Artificial Analysis reports Carbon took an average of 25.6 minutes per task on its separate Intelligence Index evaluation, against a 12.6-minute median for comparable models.
For buyers, the practical choice is not simply “best model” versus “cheapest model.” Carbon’s placement behind Fable 5.2 on MKA and the Intelligence Index suggests that benchmark leadership varies by test. Teams will need to weigh accuracy, latency, token use, access conditions and their own task results before deciding whether Carbon’s lower MKA cost matters for their work.
Products Worth Considering
Amazon Kindle Scribe (3rd Generation), 64GB, 11-inch Display, Graphite, with Premium Pen
Anker SOLIX C300 DC Portable Power Station with 60W Solar Panel, 288Wh, 300W
Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite
As an affiliate, we earn on qualifying purchases.
A Specialist Model With Trade-Offs
Google describes Carbon as its top-tier model, built on Gemini 4 Argon for demanding work such as multi-step coding and debugging and open-ended scientific research. According to the model card, it 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, positioning Carbon as a specialist rather than a default replacement.
Carbon accepts text, images, audio and video, and produces text. Its context window can hold up to 1 million tokens across input and output, and users can select low, medium or high thinking levels; medium is the default. Google lists hallucinations, occasional slowness or timeouts, and higher token use as limitations. The company’s published model card also notes a March 2026 knowledge cutoff, with some subject areas potentially less current.
Carbon had previously been made available to a small group of cyber defenders through Google’s Fairwind Program, with Artificial Analysis listing October 28 as its release date. Its stronger security features remain limited to verified defenders, according to the source report. The November 10 opening is a broader paid release, not unrestricted access to every feature.
““tasks that test the limits of a model’s intelligence””
— Google DeepMind, in the Gemini 4 Carbon model card
Benchmark Limits and Open Questions
MKA was co-built by Google DeepMind, whose model appears near the top of its results. The tasks are kept private, so outside researchers cannot inspect them or independently reproduce the scores. The benchmark’s reported margin of error also means the small gap between Carbon and Fable 5.2 should not be read as a decisive separation.
The available source does not give enough detail to establish how closely MKA’s tasks match particular business workflows, or whether its per-task cost figures include all costs that customers would face in production. Carbon’s real-world reliability, speed and total token consumption across different settings also remain open questions. Google’s model card acknowledges hallucinations and occasional timeouts, while Artificial Analysis’ separate timing figures point to a possible latency trade-off.
Access and pricing are also conditional. The initial rates expire after November 30, and app and Search access is restricted to Ultra subscribers. The source material does not specify whether those availability terms or prices will change later, or how access to Carbon’s security features may expand.
Pricing Changes and Real-World Tests
The next stated milestone is December 1, 2026, when Google’s listed API rates rise from $2/$10 to $4/$20 per million input and output tokens. Developers evaluating Carbon can compare it with alternatives under the standard rates and test their own workloads before committing to broader use. The source material does not identify a later release date or additional benchmark publication from Google.
Further evidence will depend on evaluations outside MKA, especially tests that can be independently examined and repeated. Buyers will also need to measure output quality, latency and token consumption on the tasks they actually plan to run. Until those results are available, the published numbers establish Carbon’s position on specific evaluations, not a general guarantee of lower cost or better performance.
Key Questions
Who can use Gemini 4 Carbon now?
Google made it 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 the MKA benchmark?
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 does Carbon cost per MKA task?
Google DeepMind’s report gives a cost of $5.63 per task for Carbon, compared with $35.28 for Fable 5.2 on MKA. This is a benchmark-specific figure, not a guaranteed price for other workloads.
When do Carbon’s API prices change?
The introductory rates of $2 per million input tokens and $10 per million output tokens apply through November 30, 2026. From December 1, the listed rates become $4 and $20, respectively.
Is Gemini 4 Carbon the top-ranked model?
Not across the comparisons cited here. Carbon ranks second to Fable 5.2 on MKA and on Artificial Analysis’ Intelligence Index version 5.0.1 at the listed settings. Google’s model card shows Carbon leading 15 of 17 tests it selected, but it does not include Fable 5.2.
Source: ThorstenMeyerAI.com
