🔍 Read the full analysis: Gemini 4 Carbon Is Now Available: The MKA Rank And Price Explained 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 paying developers, businesses and Google AI Ultra subscribers on November 10, 2026, and published its model card. Carbon ranks second on the new MKA benchmark, behind Claude Fable 5.2, but Google DeepMind’s report puts its cost per MKA task at $5.63 versus $35.28 for Fable; the benchmark was co-built by Google and its tasks are private.
Google opened Gemini 4 Carbon to paying developers, businesses and Google AI Ultra subscribers on November 10, while publishing a model card and introducing results from a new benchmark for AI agents. Carbon placed second on MKA, behind Anthropic’s Claude Fable 5.2, but Google DeepMind’s report lists Carbon’s cost per task at $5.63, compared with $35.28 for Fable.
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, access is limited to Google AI Ultra subscribers. Google’s developer guide identifies the model as gemini-4-carbon and says it accepts text, images, audio and video, with text output and a context window of up to one million tokens for input and output combined.
Google is offering API pricing of $2 per million input tokens and $10 per million output tokens through November 30. From December 1, the listed rates double to $4 and $20 per million tokens. These are token prices; a user’s total bill depends on how many tokens a task consumes. Google says Carbon can use more tokens, particularly at higher thinking levels, and can be slow or time out at times.
The new MKA benchmark covers 200 expert-built tasks across 10 sectors, involving material such as blueprints, scans and charts rather than only plain text. In the reported results, Claude Fable 5.2 scored 56.6%, Carbon 54.3%, and Claude Opus 5.5 53.0%. The report says the top three are within the margin of error. Google DeepMind’s report gives MKA task costs of $5.63 for Carbon and $35.28 for Fable 5.2; those figures are distinct from the API’s per-token prices.
Gemini 4 Carbon Rank and Price
Google opened Carbon to paying users on November 10, 2026. It ranked second on MKA, with a lower reported task cost than Claude Fable 5.2.
MKA results
The top three scores fall within the report’s margin of error.
Who can access Carbon
- Developers through the Gemini API and Google AI Studio
- Businesses through Gemini for business
- Gemini app access is limited to Google AI Ultra subscribers
- AI Mode in Search is limited to Google AI Ultra subscribers
API pricing changes
Token rates are separate from MKA’s estimated cost per task.
What the comparison can tell you
A narrow lead gap
MKA covers 200 expert-built tasks across 10 sectors, but its task set is private.
Check total cost
Carbon can use more tokens and may be slow or time out. Your bill depends on your tasks and token use.
Carbon’s Rank Comes With a Cost
For organizations evaluating AI agents on visual and document-heavy work, the MKA result offers one comparison of performance alongside estimated task cost. Carbon did not lead the benchmark, but the reported cost gap with Fable 5.2 could matter to buyers running many tasks. The figures are benchmark-specific estimates, not a guarantee of what every customer will pay: real costs vary with prompts, outputs, model settings and workload.
Other published comparisons give a less uniform picture. Google’s model card says Carbon leads on 15 of 17 benchmarks Google selected, but that comparison uses Google’s chosen tests and does not include Fable 5.2, which was released November 9. On Artificial Analysis’s Intelligence Index v5.0.1, Carbon at its high setting scores 60, below Fable 5.2 at 62 on high and 65 at its maximum setting. The benchmarks measure different capabilities and should not be treated as a single universal ranking.
Carbon is positioned as a high-end model for demanding, multi-step tasks rather than the default for all work. Google says Gemini 4 Argon remains the better choice for most software engineering and knowledge work. That framing, together with Carbon’s higher token use and reported slower performance, makes task fit and total cost relevant alongside headline scores.
Products Worth Considering
Samsung 32 Odyssey G5 (G51F) QHD 180Hz Gaming Monitor
LG UltraGear 34G600A-B 34-Inch Curved WQHD Gaming Monitor, 160Hz
Bose Lifestyle Ultra Soundbar and Ultra Subwoofer Bundle
As an affiliate, we earn on qualifying purchases.
From Limited Testing to Paid Access
Carbon was first made available to a small group of cyber defenders through Google’s Fairwind Program about two weeks before the November 10 public paid-access announcement. Artificial Analysis lists October 28 as its release date. Google says Carbon’s stronger security features remain limited to verified defenders; opening the model to paying users does not mean those features are available to everyone.
Google describes Carbon as a model built on Gemini 4 Argon for tasks that test the limits of model capability, including complex coding and open-ended scientific research. Its model card says Carbon uses more thinking tokens than Argon to explore possible answers and run extended verification. Users can select low, medium or high thinking levels, with medium as the default. Google gives a March 2026 knowledge cutoff and cautions that information may be less current in some domains.
Google’s model-card results are reported as of October 2026 and compare Carbon with Argon, GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1. Some listed leads are narrow: Google reports 9.6% for Carbon against 9.5% for GPT-6 Astra on SWE-sweep. In other tests, the gap is wider; on GraphWalks, Google reports 96.2% for Carbon and 71.8% for GPT-6 Astra. The results reflect the selected tests, not every real-world use.
““tasks that test the limits of a model’s intelligence””
— Google, in the Gemini 4 Carbon model card
Limits of the MKA Comparison
MKA was co-developed by Google DeepMind and Artificial Analysis, and the task set is private. That means outside researchers cannot inspect the tasks or independently reproduce the reported scores from the published material. The results are useful as a reported comparison, but readers cannot fully audit how representative the benchmark is of their own work.
The report says the top three scores fall within the margin of error, so the ranking does not establish a clear performance gap between Fable 5.2, Carbon and Opus 5.5 on MKA. The supplied material does not give the margin’s exact size. It also does not establish that the reported per-task costs will hold across different settings, prompts or customer workloads.
Google’s model-card results and Artificial Analysis’s index are separate evaluations with different tests and comparison sets. The model card excludes Fable 5.2, while the current index places that model ahead of Carbon. Google’s selected benchmark results are company-published claims; they should not be read as independent verification of overall superiority. Carbon’s performance on a particular customer’s applications, including its speed and total bill, remains uncertain.
Pricing Changes on December 1
Developers and businesses can access Carbon through Google’s paid developer and business products, subject to each product’s terms. The next clearly dated change in the supplied information is December 1, 2026, when Google’s API rates are scheduled to rise from the promotional $2/$10 per million input/output tokens to $4/$20. Buyers comparing models will need to reassess costs using their own token volumes and tasks.
Further independent scrutiny of MKA would depend on information beyond the private task set described in the report, such as additional evaluations that disclose their methods. The supplied material does not state that Google plans to release MKA tasks or publish another benchmark update. For now, Carbon’s standing is best understood through the available comparisons: second on MKA, a strong showing across Google-selected tests, and a higher reported task cost than Fable 5.2 under MKA’s accounting.
Key Questions
Who can use Gemini 4 Carbon now?
Paying developers and businesses can access it through the Gemini API, Google AI Studio and Gemini for business. In the Gemini app and AI Mode in Search, Google limits access to Google AI Ultra subscribers.
How does Carbon rank 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 report says the top three are within the margin of error, and the exact margin is not provided in the supplied material.
What does Gemini 4 Carbon cost?
Google’s API price is $2 per million input tokens and $10 per million output tokens through November 30, 2026. From December 1, it is scheduled to rise to $4 and $20, respectively. Total cost depends on token use; Google says Carbon may use more tokens than Argon.
Is Carbon the top-performing model?
Not across the comparisons provided. Google says Carbon leads 15 of 17 tests in its model card, but that is a Google-selected set that does not include Claude Fable 5.2. Carbon ranks second on MKA and scores below Fable 5.2 on Artificial Analysis’s Intelligence Index v5.0.1.
Can outsiders verify the MKA results?
Not fully from the described material. Google DeepMind co-built MKA with Artificial Analysis, and its expert-built tasks are private, preventing outside researchers from inspecting or reproducing the benchmark directly.
Source: ThorstenMeyerAI.com
