🔍 Read the full analysis: Making Sense Of Claude Fable 5.2 And Gemini 4 For Work on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Artificial Analysis’s November 17 comparison ranks Claude Fable 5.2 above Google’s Gemini 4 models on its overall Intelligence Index, but individual task results vary. Gemini 4 Carbon leads several work-related tests and costs less per benchmark task; Fable 5.2 scores higher on coding and long-document reasoning. Reliability and pricing also differ, and Google’s introductory Carbon and Argon prices are due to double December 1.
Claude Fable 5.2 leads the overall score in a November 17 comparison of Anthropic and Google’s latest AI models, but Gemini 4 Carbon outperforms it on several business and technical tests and costs less per benchmark task. The results, drawn largely from Artificial Analysis evaluations, offer no single answer for employers choosing a model: performance varies by task, while costs and the models’ tendency to guess when uncertain differ substantially.
Anthropic released Fable 5.2 on November 9. Google opened Gemini 4 Carbon to paying users on November 10, then released the faster Gemini 4 Hydrogen to all users, including through Google Search, on November 12. Gemini 4 Argon is the family’s efficient model. Artificial Analysis’s Intelligence Index, accessed November 17, gives Fable 5.2 a score of 65 at Max effort and 62 at High. Carbon scores 60 at High, Hydrogen 54 at its highest setting and Argon 50 at High.
The index’s estimated cost per task is $26.54 for Fable 5.2 at Max and $9.08 at High, compared with $5.48 for Carbon, $4.97 for Argon and $3.16 for Hydrogen. These are benchmark-task costs, not a fixed price for every workplace request. Google’s Carbon and Argon estimates use introductory prices that are scheduled to double on December 1. Artificial Analysis says one reason token rates alone do not settle the bill is that models can use different numbers of tokens while reasoning.
Results shift across individual evaluations. At the High setting, Fable 5.2 scores 49.1% on FrontierCode-AA 2.0, a test of coding in open-source repositories, against Carbon’s 40.6%. Carbon leads on Terminal-Bench 4.0, at 65.2% to 59.6%; on GDP.pdf, a professional-document reasoning test, at 28.4% to 21.8%; and on AutomationBench-AA business workflows, at 77.8% to 71.4%. Fable 5.2 leads on the AA-LCR long-document test, at 83.3% to 79.3%, and on the AA-Briefcase office-style knowledge-work rating, with an Elo score of 1,677 to Carbon’s 1,611.
Choosing by Task and Cost
For companies, the comparison suggests that choosing by overall leaderboard position alone can miss the demands of a specific job. Fable 5.2’s higher overall index score does not mean it leads every work test. Carbon’s results are stronger in the reported terminal, professional-document and business-automation evaluations, while Fable 5.2 leads in the cited coding and long-context tests. Those differences may matter when teams are deciding which model to route to particular workflows.
Cost is another practical distinction. In Artificial Analysis’s benchmark, Hydrogen has the lowest listed cost per task, while Fable 5.2 at Max is the most expensive of the models compared. The figures are tied to the test and pricing assumptions, however; they should not be treated as predictions of a team’s total bill. Google’s scheduled December 1 change will also make Carbon and Argon less cheap relative to the introductory figures used in the comparison.
Reliability results add a separate trade-off. The source reports Fable 5.2 answering more questions correctly overall, but Carbon being less likely to give an incorrect answer when it does not know. That distinction can affect how much checking a workflow needs, especially when an incorrect response could be passed on as fact. The reported figures are test results, not guarantees about performance in every setting.
Popular right now

KitchenAid Artisan 5-Quart Tilt-Head Stand Mixer with Flex Edge Beater

Dyson Airwrap Co-anda2x Multi-Styler and Dryer (2026 Model)

Vitamix 5200 Blender with 64 oz Container, Black
As an affiliate, we earn on qualifying purchases.
Three Gemini Models, One Comparator
The comparison covers different products and configurations, rather than two identical models tested at every possible setting. Fable 5.2 offers five effort levels from Low to Max; the source lists a one-million-token context window and up to 128,000 output tokens. Gemini 4 is a family: Carbon is described by Google as its most capable reasoning model, Argon as its efficient frontier model, and Hydrogen as designed for rapid execution across agentic workflows. All three Gemini models also have a one-million-token context window, according to the source.
Artificial Analysis provides the independent evaluations cited here, while each company also publishes its own results. Anthropic’s launch table reports Fable 5.2 ahead of Carbon on FrontierSWE v2 and OSWorld 2.1 computer use, while Carbon is ahead on AutomationBench. Google says Carbon ranks first on CursorBench 4.0, Perplexity’s WANDR and Zapier’s AutomationBench at more than twice lower cost per task. Those company-reported results are vendor runs that, according to the source, outsiders cannot repeat.
A second comparison from Epoch AI puts Fable 5.2 at 173.44 and Carbon at 169.76 on its Capabilities Index. The reported uncertainty ranges are 169.93–177.49 for Fable and 166.48–173.5 for Carbon; they overlap. The source describes that result as consistent with the ordering, but the overlapping ranges are a reason not to treat the small score gap as conclusive.
““the most ambitious, open-ended technical work””
— Anthropic, describing Claude Fable 5.2
Limits of the Benchmark Results
The figures do not establish which model will work best for a particular organization. Artificial Analysis’s scores reflect specific tests, model settings and access conditions; actual results can depend on prompts, connected tools, task complexity and how employees review outputs. The source does not provide enough detail to calculate a typical company’s monthly bill from the benchmark-task costs.
The reliability comparison also needs careful interpretation. The source reports 77.2% accuracy for Fable 5.2 and 58.3% for Carbon, alongside hallucination rates of 58.0% and 11.4%, respectively. It does not supply further methodological detail in the material provided here to explain how those measures are defined or how they should be translated into workplace error rates. Neither metric shows what will happen on every real-world task.
Google’s Carbon and Argon prices are set to rise on December 1, but the comparison provides no later pricing data. The source also notes that Anthropic’s Claude Mythos 5.2, described as a version of Fable 5.2 without some safety blocks, is restricted to organizations accepted into the company’s cyber and life-sciences verification programs. It does not provide further details about access criteria or availability.
Pricing Changes and Further Testing
The next listed pricing milestone is December 1, 2026, when Google’s introductory Gemini 4 Carbon and Argon token prices are scheduled to double, according to the Gemini API pricing page cited in the source. Teams comparing costs will need to recalculate using the new rates rather than rely on the introductory figures behind the November 17 task-cost comparison.
For now, the source’s practical conclusion is to match models to the work being evaluated and check the results under the organization’s own conditions. Further independent evaluations and post-price-change cost comparisons could clarify whether the current benchmark advantages hold across additional tasks. No new comparative test or company response is specified in the supplied material.
Key Questions
Which model leads the overall comparison?
Claude Fable 5.2 leads Artificial Analysis’s Intelligence Index in the cited results, scoring 65 at Max effort and 62 at High. Gemini 4 Carbon scores 60 at High, but individual test leaders vary.
Is Gemini 4 Carbon cheaper than Fable 5.2?
In Artificial Analysis’s reported benchmark, Carbon costs $5.48 per task, versus $9.08 for Fable 5.2 at High. The comparison uses Google introductory prices scheduled to double on December 1, and benchmark costs are not a forecast of every company’s bill.
Which model performs better for coding?
Fable 5.2 scores 49.1% to Carbon’s 40.6% on FrontierCode-AA 2.0, which tests coding in open-source repositories. Carbon leads on the separate Terminal-Bench 4.0 evaluation, so results depend on the kind of technical work being tested.
What do the reported reliability figures show?
The source reports higher accuracy for Fable 5.2, at 77.2% versus Carbon’s 58.3%, but a higher hallucination rate when it does not know an answer: 58.0% versus 11.4%. These are test results; the supplied material does not establish how directly they predict workplace error rates.
When do Google’s Carbon and Argon prices change?
The introductory API prices cited for Gemini 4 Carbon and Argon are scheduled to double on December 1, 2026. The comparison was accessed November 17 and uses the lower introductory rates for its cost-per-task estimates.
Source: ThorstenMeyerAI.com
