WeirdML leaderboard
As of 15 Nov 2026, the top score on WeirdML is 93.6% by GPT-6 Astra (OpenAI), ahead of Claude Fable 5.1 at 92.9%, across 159 model configurations with a published result (WeirdML, result dated 3 Sep 2026). WeirdML poses unusual machine-learning problems, such as classifying odd synthetic data or learning from strange input formats, that cannot be solved with a memorised recipe. The model must write training code that actually works on held-out data. WeirdML counts toward the BenchLeader coding score.
Key facts (as of 15 Nov 2026)
- GPT-6 Astra holds the top WeirdML score at 93.6% (WeirdML, 3 Sep 2026).
- Claude Fable 5.1 is second on WeirdML at 92.9% (WeirdML, 1 Sep 2026).
- WeirdML: Unusual machine-learning tasks that require writing working training code (WeirdML).
- 159 model configurations have a published WeirdML result on BenchLeader (BenchLeader, 15 Nov 2026).
- WeirdML feeds the coding category of the BenchLeader Index (BenchLeader methodology).
Unusual machine-learning tasks that require writing working training code.
- Published by
- WeirdMLdata via Epoch AI Benchmarking Hub
- Category
- Coding
- Index weight
- 0.5
- Models
- 159
- Data as of
- 15 Nov 2026
CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.
What the test looks like
WeirdML poses unusual machine-learning problems, such as classifying odd synthetic data or learning from strange input formats, that cannot be solved with a memorised recipe. The model must write training code that actually works on held-out data.
How it is scored
Accuracy achieved by the model's own code on the hidden test set, averaged across tasks; run by the benchmark's author.
What to keep in mind
Small task set, and the score blends coding skill with ML judgment. Run-to-run variance is noticeable because training runs themselves vary.
- 1GPT-6 Astra (pro)93.6%
- 2GPT-6 Astra (max)93.3%
- 3Claude Fable 5.1 (max)92.9%
- 4GPT-6 Astra (high)92.9%
- 5Claude Fable 5.1 (high)92.3%
- 6Claude Fable 5 (max)91.9%
- 7Claude Opus 5 (max)91.8%
- 8Claude Opus 5 (high)91.6%
- 9GPT-5.6 Sol (pro)89.4%
- 10GPT-5.6 Sol (high)88.8%
- 11Claude Fable 5 (high)87.8%
- 12GPT-5.6 Sol (max)87.0%
- 13Claude Opus 586.3%
- 14GPT-5.5 (xhigh)84.9%
- 15Gemini 3.8 Flash (high)84.8%
| # | ||||
|---|---|---|---|---|
| 1 | 93.6% | – | 2026-09-03 | |
| 2 | 93.3% | 67.3 | 2026-09-03 | |
| 3 | 92.9% | 66.7 | 2026-09-01 | |
| 4 | 92.9% | 66.3 | 2026-09-03 | |
| 5 | 92.3% | 63.7 | 2026-09-01 | |
| 6 | 91.9% | 65.1 | 2026-06-09 | |
| 7 | 91.8% | 65.0 | 2026-07-24 | |
| 8 | 91.6% | 64.4 | 2026-07-24 | |
| 9 | 89.4% | – | 2026-07-09 | |
| 10 | 88.8% | 63.4 | 2026-07-09 | |
| 11 | 87.8% | 63.9 | 2026-06-09 | |
| 12 | 87.0% | 65.3 | 2026-07-09 | |
| 13 | 86.3% | 63.2 | 2026-07-24 | |
| 14 | 84.9% | 64.1 | 2026-04-23 | |
| 15 | 84.8% | 59.4 | 2026-09-02 | |
| 16 | 83.9% | 63.5 | 2026-04-23 | |
| 17 | 82.9% | – | 2026-05-28 | |
| 18 | 82.6% | 60.2 | 2026-07-16 | |
| 19 | 79.3% | – | 2026-02-05 | |
| 20 | 78.3% | 56.8 | 2026-07-09 | |
| 21 | 78.0% | 62.5 | 2026-02-05 | |
| 22 | 77.9% | 62.9 | 2026-02-05 | |
| 23 | 77.9% | 62.1 | 2026-02-05 | |
| 24 | 77.7% | 63.1 | 2026-03-05 | |
| 25 | 76.4% | 60.0 | 2026-04-16 | |
| 26 | 76.4% | 63.1 | 2026-04-16 | |
| 27 | 76.0% | – | 2026-05-28 | |
| 28 | 75.5% | 59.7 | 2026-04-16 | |
| 29 | 75.4% | 58.8 | 2026-08-14 | |
| 30 | 72.2% | 59.9 | 2025-12-11 | |
| 31 | 72.1% | 59.9 | 2026-02-19 | |
| 32 | 70.5% | – | 2026-05-28 | |
| 33 | 70.1% | 58.9 | 2026-06-16 | |
| 34 | 69.9% | 60.0 | 2025-11-18 | |
| 35 | 68.8% | 59.0 | 2026-06-30 | |
| 36 | 67.3% | – | 2026-06-16 | |
| 37 | 67.3% | 60.6 | 2026-08-12 | |
| 38 | 67.2% | 52.3 | 2026-04-23 | |
| 39 | 66.2% | 54.7 | 2026-08-13 | |
| 40 | 66.1% | – | 2026-02-17 | |
| 41 | 63.7% | 57.1 | 2025-11-24 | |
| 42 | 63.4% | 56.2 | 2025-12-11 | |
| 43 | 63.0% | 56.8 | 2026-07-31 | |
| 44 | 62.6% | 62.5 | 2026-05-19 | |
| 45 | 61.6% | 56.5 | 2025-12-17 | |
| 46 | 60.9% | 54.7 | 2026-07-09 | |
| 47 | 60.8% | 57.5 | 2025-11-13 | |
| 48 | 60.7% | 58.2 | 2025-08-07 | |
| 49 | 60.4% | – | 2025-10-07 | |
| 50 | 60.3% | 61.5 | 2026-08-05 | |
| 51 | 60.3% | 52.0 | 2026-03-17 | |
| 52 | 58.2% | 55.9 | 2025-06-10 | |
| 53 | 57.4% | 53.0 | 2026-03-05 | |
| 54 | 57.4% | – | 2026-03-05 | |
| 55 | 57.1% | 55.4 | 2026-04-07 | |
| 56 | 57.0% | 57.2 | 2026-07-31 | |
| 57 | 56.1% | 61.0 | 2026-07-21 | |
| 58 | 55.9% | 55.1 | 2026-04-20 | |
| 59 | 54.5% | 58.6 | 2025-09-15 | |
| 60 | 54.5% | – | 2025-09-15 |
AI coding benchmark leaderboard · All LLM benchmarks
Related benchmarks
Top head-to-heads on WeirdML
Cite as: BenchLeader, “WeirdML leaderboard”, data as of 15 Nov 2026, https://www.benchleader.com/benchmarks/weirdml.
WeirdML: questions
- What does WeirdML measure?
- WeirdML poses unusual machine-learning problems, such as classifying odd synthetic data or learning from strange input formats, that cannot be solved with a memorised recipe. The model must write training code that actually works on held-out data. Scores are reported in percent of tasks solved; higher is better.
- What is the top score on WeirdML?
- As of 15 Nov 2026, the highest score on WeirdML is 93.6% by GPT-6 Astra (result dated 3 Sep 2026), ahead of Claude Fable 5.1 at 92.9%. Scores come from WeirdML and are re-read daily.
- Which AI model leads WeirdML?
- GPT-6 Astra leads WeirdML with 93.6% as of 15 Nov 2026 (result dated 3 Sep 2026), ahead of Claude Fable 5.1 at 92.9%.
- How many models have WeirdML results?
- 159 model configurations have a WeirdML result on BenchLeader, all taken from WeirdML via Epoch AI Benchmarking Hub.
- Who runs WeirdML and how often is it updated?
- WeirdML is published by WeirdML. BenchLeader re-reads the published results once a day and records the date each result was published.
- Does WeirdML count toward the BenchLeader Index?
- Yes. WeirdML contributes to the coding category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.