reasoning benchmark
PaperBench
PaperBench is a benchmark for evaluating AI agents on their ability to replicate research papers. It tests models on complex, multi-step workflows involving code implementation, experimentation, and reproducing scientific results from academic publications.
Updated Aug 7, 2026
Published models3
Registry coverage3
MetricScore
EvidenceB
PaperBench leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AC | 93.0% | 100.0% | 3 | C | |
| 2 | MA | 63.5% | 50.0% | 3 | C | |
| 3 | MI | 52.6% | 0.0% | 3 | C |
Score distribution
Top published rows on the benchmark's original scale.
PaperBench highlights
The top published results on this benchmark's own scale.
What is PaperBench?
Definition and scoring fields from the benchmark registry.
PaperBench is a benchmark for evaluating AI agents on their ability to replicate research papers. It tests models on complex, multi-step workflows involving code implementation, experimentation, and reproducing scientific results from academic publications.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- PaperBench
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- paperbench|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about PaperBench.
Which model scores highest on PaperBench?
Qwen3.8 Max is currently ranked first with 93.0%.
What does PaperBench measure?
PaperBench is a benchmark for evaluating AI agents on their ability to replicate research papers. It tests models on complex, multi-step workflows involving code implementation, experimentation, and reproducing scientific results from academic publications.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
3 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.