🔍 Read the full analysis: The AI Trendline Question: How Claude Fable 5.2 Scores On Three Benchmarks on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
An analysis published November 12 says Anthropic’s Claude Fable 5.2 scored above projected trends on the Remote Labor Index, Epoch Capabilities Index and RoboDojo. The clearest gap is on freelance projects; the Epoch result is within its uncertainty range, and RoboDojo’s score comes from Anthropic rather than a verified leaderboard entry.
Claude Fable 5.2 scored above a projected performance trend on three benchmarks after its November 9 release, according to an analysis published November 12 by ThorstenMeyerAI.com. The largest gap was on a test of paid freelance projects; the lead on Epoch AI’s index was small and within its uncertainty range, while the robotics result relies on a score Anthropic measured itself.
The analysis compared the new model’s scores with trendlines fitted to earlier benchmark record-setters, leaving Fable 5.2 out of each fit. It reports a 14.1 percentage-point lead over the projected Remote Labor Index score, a 2.4-point lead on Epoch’s Capabilities Index and an 8.8-point lead on RoboDojo. These gaps are not equally strong evidence of a lasting change in AI progress.
On the Remote Labor Index, Fable 5.2 passed 41.3% of 240 projects, or 99 tasks, at a quality the test’s reviewers judged acceptable to a reasonable client. The fitted trend predicted 27.2% for the release date. The report estimates that the trend would take about 90 days to reach Fable’s score. The source chart it reviewed showed a similar gap of 14.2 points; the author says the difference likely reflects model dating choices.
On the Epoch Capabilities Index, Fable scored 173.4 versus a trend value of 171.1. Epoch’s reported uncertainty range for Fable, 169.9 to 177.5, extends below the trend projection. On RoboDojo, Anthropic reported a 64.9% simulated-task success rate, but the analysis says that result is not a verified leaderboard entry. The leaderboard’s higher listed score is 68.9% for Q0-Preview.
The AI Trendline Question How Claude Fable 5.2 Scores
A November 2026 analysis found scores above projected trends on three benchmarks, with different levels of evidence behind each result.
Three benchmark results
The analysis compared Fable 5.2’s scores with trends fitted to earlier benchmark record-setters.
| Benchmark | Fable 5.2 score | Evidence |
|---|---|---|
| Remote Labor Index | 41.3% | Clearest above-trend result |
| Epoch Capabilities Index | 173.4 | Uncertainty range extends below trend |
| RoboDojo | 64.9% | Anthropic-reported internal run |
What the labor score means
The test covered paid freelance projects reviewed against a client-acceptance standard.
- Passed 99 of 240 projects
- 41.3% met the review standard
- Most tested projects did not pass
- It does not establish performance across all jobs or real-world conditions
Confidence varies by test
The three gaps do not support equally firm conclusions.
Remote Labor Index
The lead stayed between 13.0 and 14.4 points under alternative trend fits tested by the author.
Epoch index
Its 169.9–177.5 uncertainty range crosses below the projected trend value of 171.1.
RoboDojo
The reported score comes from Anthropic, not a verified leaderboard entry; the listed Q0-Preview score is 68.9%.
How the trendlines were built
The analysis fitted trends to earlier record-setting models, excluded Fable 5.2, then projected to its release date.
Use benchmark entries that set new highs.
Use logistic curves for percentage tests and a straight line for Epoch.
Measure each projected score against Fable 5.2’s reported result.
The Freelance Test Shows the Largest Gap
The results matter because they offer different evidence about whether the latest model’s gains exceed the pace seen in earlier top scores. The Remote Labor Index result is the strongest of the three in this analysis: the 14.1-point gap remains between 13.0 and 14.4 points under the alternative trend-fitting choices the author tested. A rough sampling margin estimated by the author is about plus or minus six points, smaller than that gap.
That result still does not mean the model can complete most professional work successfully. 41.3% of the tested projects passed, leaving the majority below the test’s client-acceptance standard. Nor does a benchmark result establish how the model performs across all jobs, users or real-world conditions. The index covers a defined set of projects and a particular review process.
The other two findings call for more restraint. The Epoch score is only 2.4 points above the fitted line, and its uncertainty range overlaps values below that line. The RoboDojo gap varies substantially—from 3.8 to 23.3 points under reasonable fitting alternatives—and depends on an internal Anthropic run. Taken together, the analysis supports an above-trend result on these measures, but not a single, equally firm conclusion across all three.
Products Worth Considering
Apple iPad 11-inch (A16), 128GB, Wi-Fi, Silver
Kindle Paperwhite Signature Edition 32GB E-Reader, 7-Inch Display, Seaglass Green
Apple Watch Series 11 GPS + Cellular, 46mm Jet Black Aluminum Case with Black Sport Band (M/L)
As an affiliate, we earn on qualifying purchases.
How the Three Trendlines Were Built
The report’s method takes earlier models that set a new high score on each benchmark and fits a trend through those record-setters. It then projects that trend to Fable 5.2’s release date. For the two percentage-based tests, the author used a logistic curve, which accounts for the 100% ceiling; for the Epoch index, the author used a straight line. The model under review was excluded from its own trendline.
The Remote Labor Index was developed by Scale AI and the Center for AI Safety. Its 240 projects came from 358 Upwork freelancers across 23 job categories, including game development, architecture, product design and data analysis. Scale AI reports that the projects took human workers an average of 28.9 hours, with a median of 11.5 hours, and had an average value of $632.60. According to Anthropic’s system card, Scale AI ran the Claude models at high effort. The index says it has no confidence interval for its automation rate.
Epoch AI’s index combines results from more than 50 benchmarks to support comparisons across time, rather than relying on a single test that may approach its score ceiling. Epoch cautions that absolute ECI scores have no meaning by themselves, though comparisons between models can be meaningful. For the labor trend, the analysis used 11 record-setting entries with release dates in Epoch’s dataset, from Gemini 2.5 Pro in June 2025 through Opus 5.5 in September 2026. Several agent products without recorded release dates were excluded.
Confidence Varies by Benchmark
The analysis is an independent reconstruction of leaderboard data and release dates, not a peer-reviewed study. Its trend projections depend on which record-setters are included, how release dates are assigned and the shape of the fitted curve. The author reports that the Remote Labor Index gap changes little under alternative fits, but the RoboDojo gap changes considerably.
Anthropic’s RoboDojo score has not been verified as a leaderboard result in the material provided. It is also unclear from the source how its internal run compares in setup and evaluation with listed entries. The Epoch result’s uncertainty interval crosses below its projected trend. For the labor test, the leaderboard provides no confidence interval; the author’s approximate sampling margin is an estimate, not an official index measure.
The source article also discloses that it was researched and drafted with AI assistance, including Anthropic’s Claude, the subject of the report. It says Anthropic, Scale AI, Epoch AI and the RoboDojo team did not review or sponsor the article. That disclosure is relevant when weighing the analysis, but does not by itself establish whether its calculations are correct.
Independent Results Could Clarify the Picture
The next useful evidence would be independently verifiable RoboDojo results and additional Fable 5.2 evaluations using comparable test conditions. Further Remote Labor Index entries could show whether the model’s measured lead persists beyond this set of 240 projects. Updated Epoch scores and future benchmark-leading models will also provide more data for judging whether the fitted trend remains a useful comparison.
The source material does not identify a scheduled benchmark update, independent replication or response from Anthropic. Until those details emerge, the three reported gaps should be read with their different evidentiary limits: a comparatively robust lead on this labor test, a small index lead within uncertainty, and a robotics score reported by the model maker.
Key Questions
What did Claude Fable 5.2 score on the three benchmarks?
The analysis reports 41.3% on the Remote Labor Index, 173.4 on Epoch’s Capabilities Index and 64.9% on RoboDojo, the last figure reported by Anthropic.
Which result provides the clearest evidence of an above-trend score?
The Remote Labor Index result. The analysis places Fable 5.2 14.1 percentage points above its projected trend, and says that gap remains between 13.0 and 14.4 points under other fitting choices it tested.
Does the Epoch result clearly exceed the trend?
Not conclusively. Fable 5.2’s reported score of 173.4 is 2.4 points above the trend estimate, but Epoch’s uncertainty range for the score—169.9 to 177.5—extends below the trend value of 171.1.
Is the RoboDojo score independently verified?
Not in the source material. The 64.9% result is described as Anthropic’s own internal run, not a verified leaderboard entry. The listed Q0-Preview score of 68.9% is higher.
Does a 41.3% labor score mean Fable 5.2 can complete most freelance work?
No. It means the model’s deliverables met the test’s client-acceptance standard on 41.3% of the 240 sampled projects. Most projects did not pass, and the result applies to this benchmark’s task set and evaluation method.
Source: ThorstenMeyerAI.com
