🔍 Read the full analysis: What We Learned Rebuilding The Math On Claude Fable 5.2’S AI 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
A November 12 analysis rebuilt trendlines for Claude Fable 5.2 on three AI benchmarks and found it above each projected score, but with unequal evidence. The Remote Labor Index result is the clearest; the Epoch index lead falls within its uncertainty range, while the RoboDojo result comes from Anthropic and is sensitive to how the trend is fitted.
Claude Fable 5.2 scored above historical trend estimates on three AI benchmarks, according to a November 12 analysis by ThorstenMeyerAI.com that reconstructed the calculations from leaderboard data and Anthropic’s launch materials. The results vary in strength: its freelance-work score is well above the estimate, its lead on a composite index is within that index’s stated uncertainty, and its simulated-robotics result relies on an Anthropic-reported run rather than a verified leaderboard entry.
The analysis compared each benchmark result with a trend fitted to earlier record-setting models, leaving Fable 5.2 out of the fit. On the Remote Labor Index (RLI), Fable 5.2 completed 41.3% of 240 freelance projects at a level reviewers judged acceptable to a reasonable client. The fitted trend projected 27.2% for its November 9 release date, putting the score 14.1 percentage points above the line. The analysis estimates that the trend would take about 90 days to reach Fable 5.2’s result.
On the Epoch Capabilities Index (ECI), Fable 5.2 scored 173.4, compared with a fitted trend estimate of 171.1: a gap of 2.4 points, or about 1.7 months at the trend’s pace. But Epoch’s reported uncertainty range for the model, 169.9 to 177.5, extends below the trend estimate. The analysis therefore finds a nominal lead, not a clearly separated result.
For RoboDojo, Anthropic reported a 64.9% success rate on simulated robot tasks. ThorstenMeyerAI.com’s trend fit placed that score 8.8 percentage points above the projected result, but the reported figure was not a verified leaderboard entry. The analysis says alternative reasonable fitting choices produce gaps from 3.8 to 23.3 points. A leaderboard entry for Q0-Preview scored 68.9%, higher than the figure Anthropic reported for Fable 5.2.
Claude Fable 5.2 above the trend, with caveats
A November 12 analysis rebuilt historical trends across three benchmarks. The evidence is strongest for freelance tasks; uncertainty and verification concerns temper the other results.
Three benchmarks, different evidence
Each result exceeded its reconstructed trend, but the strength of the comparison differs.
Freelance work
Reviewers accepted work on 41.3% of 240 projects, well above the 27.2% trend estimate. The finding applies to this test sample.
Capabilities index
The 2.4-point lead sits within Epoch’s reported uncertainty range of 169.9 to 177.5.
Simulated robotics
Anthropic reported a 64.9% result. It is not a verified leaderboard entry, and the estimated trend gap depends on the curve used.
How the trends were rebuilt
The analysis used earlier record-setting models and left Fable 5.2 out of each fit.
Raw leaderboard data and Anthropic’s launch materials informed the comparisons.
The analysis used logistic curves for percentage scores and a straight line for the Epoch index.
Alternative fitting choices produced RoboDojo gaps ranging from 3.8 to 23.3 percentage points.
What the scores do and don’t show
Benchmark results reflect specific tasks, measurements and data sources.
- RLI reviewers accepted work on 99 of 240 projects.
- That result is above the reconstructed RLI trend.
- The RLI sample does not represent all freelance work.
- The Epoch lead is within stated uncertainty.
- RoboDojo needs independent verification.
The Freelance Score Leads the Evidence
The findings compare Fable 5.2’s benchmark results with historical trend estimates. Its 41.3% RLI rate is above the trend estimate, and the analysis says its estimated sampling margin would not erase the gap. The score indicates that reviewers accepted the work on 41.3% of the projects in the test; the remaining projects did not meet the acceptance threshold. The result applies to this test and does not establish how the model performs across freelance work generally.
The ECI difference is small relative to the index’s reported uncertainty range. The RoboDojo estimate varies depending on the curve used and is based on a result reported by the model developer. Benchmark comparisons can be affected by measurement uncertainty, test design and data provenance, as well as by the fitted trend.
Products Worth Considering
Bose QuietComfort Headphones (2nd Gen), Wireless Over-Ear Noise-Cancelling Headphones, Black
EcoFlow DELTA 3 Classic Solar Generator with 220W Solar Panel
Bose Smart Dolby Atmos Soundbar, All-in-One Soundbar for TV, A.I. Dialogue Mode, Voice Control and Amazon Alexa Built-in, Supports Bluetooth
As an affiliate, we earn on qualifying purchases.
How the Three Trendlines Were Built
ThorstenMeyerAI.com says it retrieved raw data from the three leaderboards and Anthropic’s launch post, then rebuilt the chart’s trend calculations. Its approach used earlier record-setters only and excluded Fable 5.2 from every fit. It used logistic curves for percentage-based scores, which are bounded at 100%, and a straight line for the ECI. The analysis also checked how alternative fitting choices changed the results.
The RLI, developed by Scale AI and the Center for AI Safety, evaluates 240 real projects drawn from work by 358 Upwork freelancers across 23 job categories. Human reviewers compare AI work with professional work and decide whether a reasonable client would accept the result. Scale AI’s figures put average project value at $632.60 and average human completion time at 28.9 hours. The source article says Scale AI ran the Claude models at high effort.
The analysis used 11 dated RLI record-setters, from Gemini 2.5 Pro in June 2025 through Opus 5.5 in September 2026. It omitted four leaderboard products without release dates in Epoch AI’s dataset. Its calculated estimate is close to, but not identical to, the chart it reviewed: 27.2% versus 27.1% for the trend projection, and a 14.1-point versus 14.2-point gap. The article attributes the small difference possibly to dating choices.
Uncertainty Varies by Benchmark
The RLI leaderboard does not publish a confidence interval for its automation rate. ThorstenMeyerAI.com estimates a sampling margin of roughly plus or minus six percentage points for a 41% result across 240 projects; this is the article’s estimate, not a published leaderboard interval. It says that range would not close the 14.1-point gap, but the result still reflects the particular project sample, evaluation criteria and model settings.
For the ECI, the analysis reports Epoch’s uncertainty range as 169.9 to 177.5, leaving open whether Fable 5.2’s score is meaningfully above the trend. For RoboDojo, the source does not provide independent verification of Anthropic’s 64.9% run. The trend-gap estimate is also method-sensitive, ranging from 3.8 to 23.3 points under alternative reasonable fits. The available material does not establish how the models’ relative performance would change under different task sets or evaluation conditions.
Independent Results Would Clarify the Picture
The analysis does not report a planned follow-up test or a new benchmark release date. Additional leaderboard results, independent confirmation of the RoboDojo score, and further RLI and ECI results would provide more information for comparing the models and assessing uncertainty.
The analysis reports that Fable 5.2 is above the reconstructed RLI trend, while its ECI lead is within the index’s stated uncertainty range and the RoboDojo comparison relies on an unverified, developer-reported result and a method-sensitive trend estimate. These findings could change as benchmark data, release-date records or independent evaluations are updated.
Key Questions
What did the analysis find about Claude Fable 5.2?
It found Fable 5.2 above reconstructed trends on all three benchmarks, with the strongest evidence on the Remote Labor Index. The Epoch index lead is small relative to its uncertainty, and the RoboDojo result has verification and trend-fitting caveats.
What does the 41.3% Remote Labor Index score mean?
Fable 5.2’s work met reviewers’ reasonable-client acceptance standard on 99 of 240 projects. It does not mean the model completed every task successfully or that the result generalizes to all freelance work.
Was Fable 5.2 above the trend on the Epoch index?
Its score of 173.4 was above the analysis’s trend estimate of 171.1. However, Epoch’s reported uncertainty range, 169.9 to 177.5, reaches below the trend estimate, so the difference is not clearly outside that range.
How reliable is the RoboDojo comparison?
It is less settled than the RLI comparison. The 64.9% result was reported by Anthropic, not listed as a verified leaderboard entry, and the calculated gap varies substantially with the trend-fitting method.
Source: ThorstenMeyerAI.com
