🔍 Read the full analysis: What Happens When You Recalculate Claude Fable 5.2’S AI Benchmark Scores? on ThorstenMeyerAI.com
Get hardware and tech essentials 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
A recalculation of benchmark data puts Claude Fable 5.2 above the projected trend on the Remote Labor Index, Epoch Capabilities Index and RoboDojo. The freelance-work result is the strongest signal; the Epoch lead falls within its uncertainty range, while RoboDojo relies on Anthropic’s own unverified measurement and varies with the trend-fitting method.
A November 12 analysis of Claude Fable 5.2 recalculated its results against past performance trends on three AI benchmarks, finding the model above all three projected scores but with unequal evidence. The clearest lead was on the Remote Labor Index, which tests real freelance projects; the Epoch Capabilities Index result was within its stated uncertainty range, and the RoboDojo result came from an Anthropic-run evaluation rather than a verified leaderboard entry.
The analysis, published by ThorstenMeyerAI.com after Anthropic released Fable 5.2 on November 9, compared the model’s scores with trendlines built from earlier record-setting systems. Its author says he retrieved leaderboard data and Anthropic’s launch material, then fitted the trends with Fable 5.2 excluded. For percentage-based benchmarks, he used logistic curves; for the Epoch index, he used a straight line. The result is an assessment of performance relative to earlier progress, not a new benchmark test.
On the Remote Labor Index (RLI), Fable 5.2 completed 41.3% of 240 projects to a standard reviewers judged a reasonable client would accept. The trendline projected 27.2% for the model’s November 9 release date, leaving a 14.1 percentage-point gap. The analysis estimates the model reached that score roughly 90 days ahead of the trend. It says alternative fitting choices leave the RLI gap between 13.0 and 14.4 points.
The other results are less conclusive. Fable 5.2 scored 173.4 on the Epoch Capabilities Index, compared with a projected 171.1, a 2.4-point lead the analysis translates to about 1.7 months. Epoch AI’s uncertainty range for the model, 169.9 to 177.5, extends below the trend. On RoboDojo, Anthropic reported a 64.9% success rate, 8.8 points above the analysis’s trend fit. The author says other reasonable fitting methods produce gaps from 3.8 to 23.3 points, and a leaderboard entry for Q0-Preview is higher at 68.9%.
Claude Fable 5.2’s scores above the trend?
A recalculation found above-trend scores on three benchmarks, but the strength of the evidence varies. The comparisons measure progress against past results, not a new benchmark test.
Evidence differs by benchmark
The recalculated scores sit above their projected trends, but only one result is a consistent lead across the reported fitting methods.
Freelance project work
Fable 5.2 completed 41.3% of 240 projects; alternative trend fits leave a 13.0–14.4 point gap.
Epoch index
Its 173.4 score is above the 171.1 trend, but the stated uncertainty range extends below it.
Robot control
The 64.9% result is from Anthropic’s own evaluation; estimated lead varies from 3.8 to 23.3 points across fitting methods.
What “above trend” means
The analysis compared Fable 5.2 with curves fitted to earlier record-setting systems.
Exclude Fable 5.2 and fit a trend to earlier benchmark record-setters.
Estimate the score the past trend would predict for its November 9 release.
Measure the gap between the reported score and that projection.
What readers should take from it
The results show progress relative to prior systems, with limits on what each comparison can establish.
- A steady above-trend result on the tested freelance projects.
- Above-projection scores on all three benchmarks.
- Whether the Epoch lead exceeds its uncertainty.
- Whether RoboDojo’s reported result is independently verified.
- How well the project result generalizes to all freelance work.
The Freelance Benchmark Shows the Clearest Lead
The results indicate that Fable 5.2 scored above reconstructed trendlines on three benchmarks, but the comparisons differ in evidentiary strength. The RLI result is the most consistent across the fitting methods reported: its estimated gap ranges from 13.0 to 14.4 percentage points. The author estimates a sampling margin of about six percentage points for a 41% rate across 240 projects.
Fable 5.2 passed fewer than half of the projects: 41.3% means most tasks did not meet the stated client-acceptance standard. The result applies to this set of projects and evaluation process; it does not measure all freelance work or establish a general level of job replacement. The analysis also notes that the leaderboard provides no confidence interval for its automation rate.
The Epoch gap is not clearly distinguishable from the index’s stated uncertainty range. The RoboDojo estimate comes from an Anthropic-run evaluation, and the estimated gap varies substantially with the trend-fitting method. The analysis therefore reports a consistent above-trend result on the RLI, an Epoch result within the stated uncertainty range, and a RoboDojo comparison with methodological and verification limitations.
Products Worth Considering
Logitech MX Master 4 for Mac Wireless Ergonomic Mouse, White Silver
Logitech MX Ergo S Advanced Wireless Trackball Mouse USB-C – Graphite
Apple MacBook Pro 16-inch (2021), M1 Pro, 10-core CPU, 16-core GPU, 16GB RAM, 512GB SSD, Silver
As an affiliate, we earn on qualifying purchases.
How the Three Trendlines Were Rebuilt
The analysis defines “above trendline” by comparing a new model with a curve fitted to earlier benchmark record-setters. It excludes Fable 5.2 from each fit, projects the curve to the model’s release date, and measures the score difference. It also estimates how long the earlier trend would take to reach Fable 5.2’s score. Those calculations depend on the chosen models, dates and curve shape; they are not official benchmark ratings.
The Remote Labor Index, developed by Scale AI and the Center for AI Safety, covers 240 projects drawn from work by 358 Upwork freelancers across 23 categories, including game development, product design, architecture and data analysis. Scale AI reports an average human completion time of 28.9 hours and average project value of $632.60. Reviewers compare AI deliverables with professional work, counting a project as passing when the work is at least as good as what a reasonable client would accept. Anthropic’s system card says Scale AI ran its Claude models at high effort.
For the RLI trend, the author used 11 record-setters, from Gemini 2.5 Pro in June 2025 through Claude Opus 5.5 in September 2026. Four agent products without release dates in Epoch AI’s dataset were excluded. The source chart’s projected score was 27.1%, versus the author’s 27.2%; the article says the small difference likely reflects dating choices. The Epoch index combines results from more than 50 benchmarks, while RoboDojo tests simulated robot-arm control. These measure different abilities, so their trend comparisons should not be treated as interchangeable.
“Absolute ECI values are meaningless by themselves, but meaningful comparisons can be made between models.”
— Epoch AI, describing the Epoch Capabilities Index
Confidence Varies Across the Benchmarks
The reported calculations are an independent analysis, not a confirmation by the benchmark organizations. The Epoch score’s stated range of 169.9 to 177.5 overlaps the trend value of 171.1, so the available figures do not establish a clear lead on that index. For RoboDojo, the 64.9% figure is attributed to Anthropic’s internal run; the source material does not provide an independent verification or explain whether its setup is directly comparable with every leaderboard result.
Trend estimates also depend on which record-setting models are included and how their release dates and scores are fitted. That sensitivity is particularly wide for RoboDojo. The RLI result is steadier under the alternative fits reported, but its leaderboard has no confidence interval, and the article’s approximate sampling estimate is not an official uncertainty measure. The source material does not establish whether the three evaluations used identical settings across competing systems or whether later audits will change any scores.
Independent Checks Could Clarify the Gaps
Potential sources of further evidence include independent verification of the RoboDojo result, fuller documentation of the evaluation setup, and additional leaderboard entries. For the RLI, more runs or a published uncertainty estimate could help assess how precisely the 41.3% result applies beyond the 240 tested projects. Updated Epoch uncertainty estimates or additional model scores could help determine whether the index gap persists.
The supplied material does not confirm further testing or review. Based on the reported figures, Fable 5.2 is above the reconstructed RLI trend, its Epoch score is above the trend estimate but within the stated uncertainty range, and its RoboDojo score is above the fitted trend by an amount that varies with the method and is based on Anthropic’s report.
Key Questions
What does “above the trendline” mean in this analysis?
It means Fable 5.2’s score is higher than a projection based on earlier benchmark record-setters at the model’s release date. The analysis excludes Fable 5.2 when fitting each trend, but the result still depends on the models, dates and curve chosen.
Which benchmark showed the strongest result?
The Remote Labor Index showed the clearest lead: 41.3% against a projected 27.2%, or 14.1 percentage points above the fitted trend. The analysis says that gap stays between 13.0 and 14.4 points under other fitting choices it tested.
Does the Epoch result establish a clear lead?
No. Fable 5.2’s 173.4 score is above the 171.1 trend estimate, but Epoch AI’s reported uncertainty range of 169.9 to 177.5 extends below that estimate.
Was the RoboDojo score independently verified?
The supplied analysis says Anthropic reported the 64.9% result from its own internal run and that it was not a verified leaderboard entry. It also reports a higher leaderboard score of 68.9% for Q0-Preview.
Does a 41.3% RLI score mean Fable 5.2 can complete most freelance work?
No. It passed 41.3% of the 240 tested projects under the benchmark’s client-acceptance standard, meaning most projects in that test did not pass. The result does not establish performance across all freelance jobs.
Source: ThorstenMeyerAI.com
