FML-bench
Agents propose and run experiments on eight machine learning research tasks. The authors measure task performance and how widely the proposals vary across iterations.
- Made by
- Qiran Zou and colleagues
- Released
- October 2025
- Kind
- AI research tasks
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Agentic
- Environment
- Code sandbox
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Agents
- Strongest model tested
- GPT-5 #71 on the capability index
- Last model added
- August 2025
- Scores refreshed
- November 20th, 2026
2 models share the top score of 0.50 among models released by August 2025
Best score per model from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper reports no overall score. The headline shows Generalization, the first task in Table 2. Where a model ran with several settings or harnesses, we keep its best.
The newest model with a score here came out in August 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1 Claude Opus 4.1 0.50
- 2 GPT-5 0.50
- 3
Gemini 2.5 Pro Jun 2025 0.33
Source: The FML-bench paper
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.1 | 0.50 | Claude Code | Aug 2025 | Official leaderboard |
| 2 | GPT-5 | 0.50 | TheAIScientist | Aug 2025 | Official leaderboard |
| 3 | | 0.33 | TheAIScientist | Jun 2025 | Official leaderboard |
Rankings by part
We read 8 parts from the benchmark’s paper, each ranking models on one part of the test, such as Generalization, Data efficiency and Representation learning. Gemini 2.5 Pro leads 6 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Generalization
3 models we track have a score here.
- 1 Claude Opus 4.1 0.50
- 2 GPT-5 0.50
- 3
Gemini 2.5 Pro Jun 2025 0.33
Data efficiency
3 models we track have a score here.
- 1
Gemini 2.5 Pro Jun 2025 ↑2 0.82
- 2 GPT-5 0.77
- 3 Claude Opus 4.1 ↓2 0.66
Representation learning
3 models we track have a score here.
- 1
Gemini 2.5 Pro Jun 2025 ↑2 0.86
- 2 GPT-5 0.85
- 3 Claude Opus 4.1 ↓2 0.77
Continual learning
3 models we track have a score here.
- 1
Gemini 2.5 Pro Jun 2025 ↑2 0.78
- 2 GPT-5 0.44
- 3 Claude Opus 4.1 ↓2 0.23
Causality
3 models we track have a score here.
- 1
Gemini 2.5 Pro Jun 2025 ↑2 0.95
- 2 GPT-5 0.97
- 3 Claude Opus 4.1 ↓2 0.98
Robustness and reliability
3 models we track have a score here.
- 1
Gemini 2.5 Pro Jun 2025 ↑2 0.96
- 2 GPT-5 0.93
- 3 Claude Opus 4.1 ↓2 0.89
Privacy
3 models we track have a score here.
- 1
Gemini 2.5 Pro Jun 2025 ↑2 0.17
- 2 GPT-5 0.49
- 3 Claude Opus 4.1 ↓2 0.49
Fairness and bias
3 models we track have a score here.
- 1 GPT-5 ↑1 0.04
- 2
Gemini 2.5 Pro Jun 2025 ↑1 0.09
- 3 Claude Opus 4.1 ↓2 0.38
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The FML-bench paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official FML-bench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in AI research
- PaperBench In use Agent replicates ICML 2024 papers from scratch. LLM judge grades against author-written hierarchical rubrics.
Qwen 3.8 Max 93%
- MLS-Bench In use Agent must invent one algorithmic change in an ML research scaffold whose gain transfers across settings. Scored relative to human baselines. Claude Fable 5.1 50.3%
- PostTrainBench 1.2 In use CLI agent gets a base model, one H100 and a time budget to post-train it for a target benchmark. Scored on normalized gain. Claude Fable 5.1 44.6%
- MLE-bench In use Kaggle-style ML engineering competitions. Agent trains models and is scored against competition leaderboards. GPT-6.2 Astra 96.1%
- PMPP-Hard In use Agents write GPU kernels for 69 tasks that must be correct, and on 48 of them also run within 1.25 times the reference time. The score is the area under the solved-tasks curve over a 10-hour budget, out of 690 task-hours. Kimi K3 0.71
- KernelBench In use Replace PyTorch operators with faster custom GPU kernels. Scored on correctness and speedup.
MiniMax-M3 28.8%
Data audit
7 of 8 checks pass, 1 warning. Look first at the Newest score check. The newest came 15 months ago, in August 2025, for GPT-5. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 1 pass 3 notes
- Pass Independent runs Scores from the benchmark’s paper (3 entries).
- Note Models scored Only 3 models we track.
- Warning Newest score The newest came 15 months ago, in August 2025, for GPT-5.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5 (#71).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 5 rows for models we track at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.