MATH500
500-problem maths subset
mathactive
Score in %
Definition
- Version
- not in the notes
- Category
- math
- Maintainer
- not in the notes
- Status
- active
- Home page
- not in the notes
Sources
- huggingface.co/LiquidAI/LFM2.5-8B-A1B https://huggingface.co/LiquidAI/LFM2.5-8B-A1B
- huggingface.co/microsoft/Phi-4-mini-flash-reasoning https://huggingface.co/microsoft/Phi-4-mini-flash-reasoning
- LIMO: Less is More for Reasoning arxiv.org https://arxiv.org/abs/2502.03387
Scores at a Glance
Score in %
Higher is better for this unit.
1
Falcon-H1R-7BTechnology Innovation Institute (TII)
LIMOShanghai Jiao Tong University
Phi-4-mini-flash-reasoningMicrosoft AI
QwQ-32B-PreviewAlibaba (Qwen)
LFM2.5-8B-A1BLiquid AI
Falcon-H1R-7BTechnology Innovation Institute (TII)97.4 %
vendor-reported2LIMOShanghai Jiao Tong University95.6 %
vendor-reported3Phi-4-mini-flash-reasoningMicrosoft AI92.45 %
vendor-reported4QwQ-32B-PreviewAlibaba (Qwen)90.6 %
vendor-reported5LFM2.5-8B-A1BLiquid AI88.76 %
vendor-reported0255075100
Where this comes from
Each bar is one model, with its most independent recorded score. Hatched bars are vendor-reported; the word beside each bar says who ran it. Settings and dates differ between rows; the Scores table below has every result.
Scores
| Model | Score | Who ran it | Setting | View or population | Date | Source |
|---|---|---|---|---|---|---|
| Falcon-H1R-7B | 97.4 % | vendor-reported | – | – | – | huggingface.co |
 Llama 3.1 Nemotron Ultra 253B v1 | 97 % | vendor-reported | Reasoning on, pass@1 | Nemotron Ultra model card | Apr 2025 | huggingface.co |
| Llama 3.3 Nemotron Super 49B v1 | 96.6 % | vendor-reported | Reasoning on, pass@1 | Nemotron Super model card | Mar 2025 | huggingface.co |
| LIMO | 95.6 % | vendor-reported | – | – | Feb 2025 | arxiv.org |
| Phi-4-mini-flash-reasoning | 92.45 % | vendor-reported | – | – | – | huggingface.co |
| QwQ-32B-Preview | 90.6 % | vendor-reported | – | – | 28 Nov 2024 | qwenlm.github.io |
| LFM2.5-8B-A1B | 88.76 % | vendor-reported | – | – | – | huggingface.co |
 DeepSeek-V2.5 (Dec 2024, DeepSeek-V2.5-1210) | 82.8 % | vendor-reported | Up from 74.8 for DeepSeek-V2.5 | DeepSeek-V2.5-1210 model card | Dec 2024 | huggingface.co |
Where this comes from
Scores from different settings, harnesses or dates are not directly comparable; the row says which. Sorted by score, highest first, as a reading aid, not a ranking.
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "MATH500", https://aurion.semai.id/benchmarks/math-500/, records of October 2026.