Claude Fable 5.2 And Gemini 4: A Work-Focused AI Showdown
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Claude Fable 5.2 And Gemini 4: A Work-Focused AI Showdown on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the little things that make your day delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

Anthropic released Claude Fable 5.2 on November 9, and Google opened Gemini 4 Carbon to paying users on November 10 before releasing Gemini 4 Hydrogen on November 12. Artificial Analysis ranks Fable 5.2 highest overall, but its tests show Gemini models leading several work tasks and costing less per benchmark task; accuracy and error rates also differ.

Anthropic’s Claude Fable 5.2 and Google’s Gemini 4 models are drawing a mixed verdict from published benchmarks: Fable leads the Artificial Analysis Intelligence Index, while Gemini 4 Carbon leads several individual work tests and costs less per benchmark task. The results, accessed on November 17, do not identify one model as best for every job, leaving teams to weigh performance, reliability and price.

Anthropic released Claude Fable 5.2 on November 9. Google made Gemini 4 Carbon available to paying users the next day, then released the faster Gemini 4 Hydrogen on November 12, including in Google Search. Gemini 4 is a family: Google describes Carbon as its most capable reasoning model and recommends the more efficient Argon for most work. Hydrogen is designed for rapid execution of agentic workflows.

On the Artificial Analysis Intelligence Index, Fable 5.2 scored 65 at Max effort and 62 at High. Carbon scored 60 at High, Hydrogen 54 at its highest setting, and Argon 50 at High. The ranking is an overall index result, not a guarantee that Fable will perform best on a particular company’s tasks.

Cost per index task also varied. Artificial Analysis put Fable 5.2 at $26.54 at Max and $9.08 at High, compared with $5.48 for Carbon, $4.97 for Argon and $3.16 for Hydrogen. Google’s Carbon and Argon task-cost figures use introductory prices; the company’s API pricing page said their token prices would double on December 1, from $2 to $4 per million input tokens and from $10 to $20 per million output tokens. Hydrogen was listed at $1.50 per million input tokens and $7.50 per million output tokens. Fable 5.2 was listed at $8 and $40, respectively. These prices do not by themselves determine a task’s final cost, since the number of tokens used also matters.

At a glance
reportWhen: Models released November 9-12, 2026; be…
The developmentNew independent benchmark results compare Anthropic’s Claude Fable 5.2 with Google’s Gemini 4 models on work tasks, costs and reliability.

Choosing Models for Different Work

The comparison matters to companies deciding whether to deploy one model broadly or select tools task by task. In the independent tests cited by ThorstenMeyerAI.com, Fable 5.2 led on coding in open-source repositories and long-document reasoning. Carbon led on terminal tasks, professional documents and business workflow automation. Those differences could affect both the quality of work delivered and the amount of human review needed.

The figures also underline a budget trade-off. Carbon and Hydrogen had lower reported costs per index task than Fable 5.2, but an overall score or benchmark price does not settle whether a model is suitable for a specific workflow. Businesses would need to test their own tasks, account for token use and check how often outputs require correction before estimating their actual costs.

Reliability is another part of that decision. In the source’s account of Artificial Analysis results, Fable had higher answer accuracy than Carbon, but also a higher rate of incorrect confident answers when it did not know. That makes the distinction between answering correctly and admitting uncertainty relevant for work where errors carry financial, legal or operational consequences.

Products Worth Considering

Therabody Theragun Prime Plus Heated Deep Tissue Massage Gun, Black

Therabody Theragun Prime Plus Heated Deep Tissue Massage Gun, Black

As an affiliate, we earn on qualifying purchases.

How the Models Were Compared

The comparison draws on Artificial Analysis evaluations, which run models through common tests and publish results by task, as well as figures from Epoch AI and the companies’ own benchmark tables. The independent tests cited include FrontierCode-AA 2.0 for coding in open-source repositories, Terminal-Bench 4.0 for terminal work, GDP.pdf for professional-document reasoning, AutomationBench-AA for business software workflows and AA-LCR v1.1 for long-context reasoning.

At the same High setting, Fable scored 49.1% on FrontierCode-AA 2.0, against Carbon’s 40.6%. Carbon scored 65.2% on Terminal-Bench 4.0, against Fable’s 59.6%; on GDP.pdf it scored 28.4%, against 21.8%; and on AutomationBench-AA it scored 77.8%, against 71.4%. Fable led on AA-LCR v1.1, with 83.3% against Carbon’s 79.3%. The source says both models failed most tasks on GDP.pdf.

Epoch AI’s Capabilities Index also placed Fable above Carbon, at 173.44 against 169.76. However, the reported uncertainty ranges overlap: 169.93 to 177.49 for Fable and 166.48 to 173.5 for Carbon. The source also reports that Anthropic’s own table showed Fable ahead on FrontierSWE v2 and OSWorld computer use, while Carbon led on AutomationBench. Google cited other evaluations in which Carbon ranked first. Company-reported results are not equivalent to independent tests.

“the most ambitious, open-ended technical work”

— Anthropic, describing Claude Fable 5.2

Limits of the Benchmark Results

Benchmark scores do not establish how either model will perform on every company’s real workload. The results summarized here cover selected tests, and the source notes that both models failed most tasks on the professional-document test. Epoch AI’s overlapping uncertainty ranges also mean its reported scores do not clearly separate the models on that index.

The source reports Fable at 77.2% accuracy against Carbon’s 58.3%, but says Fable had a 58.0% hallucination rate against Carbon’s 11.4% when the models did not know an answer. Those figures describe different aspects of performance; the source does not provide further detail here about how the hallucination measure was calculated or the precise test conditions. Readers should not treat the figures as a direct prediction of error rates in a particular workplace.

Pricing is subject to change: the cited Carbon and Argon introductory rates were scheduled to double on December 1. Actual costs also depend on token consumption and task design. The comparison’s author disclosed using AI assistance, including an Anthropic model, and said neither Anthropic, Google nor Artificial Analysis reviewed or sponsored the article.

Pricing Changes and Workplace Tests

The immediate pricing milestone is December 1, when Google’s cited Carbon and Argon API token rates are scheduled to rise. Teams comparing options will need to account for those rates alongside task-level token use, rather than relying only on the listed price per million tokens or a single benchmark score.

Further benchmark updates and hands-on evaluations may clarify how the models perform across a wider range of workplace tasks. For now, the source material does not report a later release, a change in the stated prices or a definitive winner. Companies considering either system can compare them on representative tasks and track accuracy, uncertainty, review time and cost before choosing where to use them.

Key Questions

Which model scored highest on the Artificial Analysis Intelligence Index?

Claude Fable 5.2 scored highest among the listed models, with 65 at Max effort and 62 at High. Gemini 4 Carbon scored 60 at High.

Does Gemini 4 perform better on every work task?

No. In the tests cited, Carbon led on terminal work, professional documents and business workflows, while Fable led on coding in open-source repositories and long-document reasoning.

Which model had the lowest reported cost per index task?

Gemini 4 Hydrogen had the lowest listed figure, at $3.16 per index task. The source says Carbon and Argon’s figures reflect introductory prices scheduled to end on November 30.

When are Gemini 4 Carbon and Argon prices scheduled to change?

Google’s cited API pricing page listed a price increase on December 1, doubling Carbon and Argon’s rates from $2 to $4 per million input tokens and from $10 to $20 per million output tokens.

What remains uncertain about the comparison?

The benchmark results do not establish which model will work best for a specific organization. Real-world accuracy, review requirements and total cost will depend on the tasks, settings and token use involved.

Source: ThorstenMeyerAI.com

You May Also Like

AI Development Accelerates As Three Gates Close In Just Over Two Weeks

China, the EU, and the US implement significant AI pre-release regulations within 19 days, shaping global AI deployment standards.

What Makes AI a 2026 Tech Leader? Top 10 Innovations

Explore the top 10 AI innovations shaping 2026 as a year of technological leadership, with confirmed developments and key impacts for the industry.

ByteDance R&D Head Leaves GenAI To Join The Embodied Intelligence Sector

A 36Kr headline reports a ByteDance GenAI R&D leader is leaving for embodied intelligence and says three core members departed in four months.

Phone Footage Brings Basketball Highlights Within Reach

An IdeaNavigator AI concept proposes turning phone-recorded basketball games into player highlights and stats, with validation still needed.