OpenMath-Mistral-7B-v0.1-hf: benchmark results
NVIDIA's Hugging Face-format OpenMath derivative of Mistral-7B-v0.1, trained on OpenMathInstruct-1's 1.8 million problem-solution pairs. It combines mathematical text reasoning with Python execution; problem-solution pairs are not a training-token count. Provider: NVIDIA. Access: Open.
Aggregate Rating 1375 ± 35, rank #6984 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - WinoGrande | 77.27 | Accuracy (%) (5-shot) | 54.6 |
| Open LLM Leaderboard v1 - HellaSwag | 81.78 | Normalized accuracy (%) (10-shot) | 47 |
| Open LLM Leaderboard v1 - MMLU | 59.34 | Accuracy (%) (5-shot) | 44.9 |
| Open LLM Leaderboard v1 - ARC Challenge | 59.39 | Normalized accuracy (%) (25-shot) | 43.2 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 46.13 | MC2 (%) (0-shot) | 33.5 |
| Open LLM Leaderboard v1 - GSM8K | 0.08 | Accuracy (%) (5-shot) | 9.5 |
Interactive version: theaggregate.ai/model?slug=openmath-mistral-7b-v0-1-hf · How It Works · Data refreshed daily, snapshot 2026-11-08.