AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

Anthropic released Claude Fable 5.2 on Nov. 9, 2026, and Google opened Gemini 4 Carbon to paying users the next day, followed by Gemini 4 Hydrogen on Nov. 12. Artificial Analysis scores put Fable ahead on its overall index, while Gemini models lead several task tests and cost less per task; the results do not identify one best model for every workplace.

Anthropic released Claude Fable 5.2 on Nov. 9, and Google made Gemini 4 Carbon available to paying users on Nov. 10, setting up a fresh comparison for teams choosing AI tools. Artificial Analysis data accessed Nov. 17 put Fable ahead on the overall Intelligence Index, while Gemini models led several individual work tests and had lower reported costs per task.

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

At the High effort setting, Fable 5.2 scored 62 on the Artificial Analysis Intelligence Index, compared with 60 for Gemini 4 Carbon. Fable scored 65 at Max effort, while Gemini 4 Hydrogen scored 54 at its highest setting and Gemini 4 Argon scored 50 at High. These figures summarize performance across the index; they do not establish which model will perform best on a particular company’s work.

In the task comparisons at High, Fable led on coding in open-source repositories (49.1% to Carbon’s 40.6%) and long-context reasoning (83.3% to 79.3%). Carbon led on terminal tasks (65.2% to 59.6%), professional documents (28.4% to 21.8%) and business software automation (77.8% to 71.4%).

Artificial Analysis estimated the cost of an index task at $9.08 for Fable 5.2 at High and $5.48 for Carbon. At their listed configurations, it estimated $26.54 for Fable at Max, $4.97 for Argon and $3.16 for Hydrogen. Google’s Carbon and Argon task-cost figures use introductory API prices that double on Dec. 1; task costs also reflect the tokens a model uses, not only the per-token price.

At a glance
reportWhen: Models released Nov. 9-12, 2026; benchm…
The developmentNew releases from Anthropic and Google have produced competing models whose independent benchmark results point to different strengths, costs and reliability trade-offs for work.
Bar chart titled The top score costs the most per task. Artificial Analysis Intelligence Index score and cost per index task: Claude Fable 5.2 (Max) 65 and 26.54 dollars; Claude Fable 5.2 (High) 62 and 9.08 dollars; Gemini 4 Carbon (High) 60 and 5.48 dollars; Gemini 4 Hydrogen (Xhigh) 54 and 3.16 dollars; Gemini 4 Argon (High) 50 and 4.97 dollars. A highlighted note says Carbon and Argon are at Google's introductory price and their token prices double on December 1
Index score and cost per task on the Artificial Analysis Intelligence Index. Source: Artificial Analysis model pages, accessed November 17, 2026; price change: Google Gemini API pricing page. Chart: Thorsten Meyer AI.

How the Trade-Offs Affect Teams

The results suggest that model selection may depend on the work being handed over and the cost a team can accept. A group focused on repository coding or long documents may value Fable’s higher results in those tests, while teams handling terminal tasks, business documents or workflow automation may prefer Carbon’s results. The benchmark scores are controlled comparisons, not a guarantee of performance on a team’s own data or processes.

The reported reliability measures also differ. In Artificial Analysis’ evaluation, Fable answered correctly more often overall, with 77.2% accuracy against Carbon’s 58.3%. But its reported hallucination rate was 58.0%, compared with 11.4% for Carbon. These measures describe different aspects of the test: the results indicate that Fable got more answers right, while Carbon was less likely to give a wrong answer when it did not know. That distinction may matter in workflows where unsupported answers are costly.

Price can affect which tasks are practical to run at scale. The reported per-task estimates favor Google’s models, although the Carbon and Argon estimates rely on introductory pricing that ends Dec. 1. Businesses comparing options will need to account for their actual usage, model settings and workflow requirements.

Products Worth Considering

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

As an affiliate, we earn on qualifying purchases.

A Family of Models, Not One Matchup

The releases arrived within four days. Anthropic introduced Claude Fable 5.2 on Nov. 9. Google opened Gemini 4 Carbon to paying users on Nov. 10, then released the faster Gemini 4 Hydrogen on Nov. 12, making it available to all users, including through Google Search. Gemini 4 Argon is the family’s more efficient frontier model.

Fable 5.2 is positioned by Anthropic for ambitious, open-ended technical work. The source describes a one-million-token context window and five effort settings, from Low to Max. Google describes Carbon as its most capable reasoning model and recommends Argon for most software engineering and knowledge work. The Gemini models listed in the source have one-million-token context windows.

The comparison draws on Artificial Analysis model pages accessed Nov. 17, including the Intelligence Index and task-specific evaluations for coding, terminal work, documents, automation, long context and answer reliability. Anthropic and Google supplied launch and product information; the benchmark results are attributed to Artificial Analysis. The source says neither company nor Artificial Analysis reviewed or sponsored its comparison, and that the article was researched and drafted with AI assistance, including an Anthropic model.

““the most ambitious, open-ended technical work””

— Anthropic

Where Benchmark Results May Fall Short

The figures reflect specific Artificial Analysis tests and configurations, including selected effort settings. The source does not establish how the models compare on any particular employer’s internal documents, software, security requirements or workflows. It also does not provide enough detail here to determine how benchmark performance changes under different prompts or real-world operating conditions.

Costs per task are estimates tied to the evaluated tasks and pricing in effect when measured. Google’s introductory Carbon and Argon API prices rise on Dec. 1, and the source does not give a post-increase cost-per-task estimate. Fable’s reported higher accuracy and higher hallucination rate also leave open how the measures translate to specific uses and how teams should weigh them.

Teams Need Their Own Comparisons

The next clear pricing milestone in the source is Dec. 1, when Google’s introductory Carbon and Argon API prices double. Teams weighing the models can compare them on representative tasks, record the model settings and token use, and check the results against their own accuracy and cost requirements. The available information does not identify a universal winner or specify a subsequent benchmark release date.

Key Questions

Which model leads the overall benchmark?

Claude Fable 5.2 leads the reported Artificial Analysis Intelligence Index comparison: 62 at High, versus 60 for Gemini 4 Carbon at High. Fable scores 65 at Max.

Which model is cheaper per task?

Artificial Analysis estimates $5.48 per index task for Carbon and $9.08 for Fable at High. Google’s Carbon estimate uses introductory pricing that doubles on Dec. 1; costs depend on the tested task and token use.

Which model performed better on coding?

Fable 5.2 scored 49.1% on repository coding at High, ahead of Carbon’s 40.6% in the reported test. Carbon scored higher on terminal work.

What do the reliability results show?

In the reported Artificial Analysis measures, Fable had higher accuracy (77.2% versus 58.3%), while Carbon had a lower hallucination rate (11.4% versus Fable’s 58.0%). The measures capture different aspects of the evaluation.

Source: Thorsten Meyer AI

You May Also Like

Accessibility issue triage board for small websites

A new triage board for accessibility issues on small websites is being tested as a workflow solution for small business owners and freelancers.

Best AI Content Generation Tools Compared

Compare ChatGPT and Jasper on writing quality, control, workflows, collaboration, and value to find the right AI content tool for your team.

10-Minute Client Risk Reports For Independent Financial Advisors

An IdeaNavigator AI concept proposes a one-page client risk report for solo and small-firm advisers; its speed and demand remain untested.

U.S. Lifts Restrictions on Anthropic’s Most Powerful A.I. Models

The U.S. government has removed restrictions on Anthropic’s most advanced AI models, potentially expanding their deployment and use in various sectors.