Hebrew-Mistral-7B: benchmark results
A Mistral-7B derivative continued-pretrained in English and Hebrew with an expanded tokenizer described as 64,000 tokens in the card; its evaluated configuration has 64,001 vocabulary entries. The card credits Yam Peleg in collaboration with Jonathan Rouach and Arjeo, inc. Provider: Other. Access: Open.
Aggregate Rating 1370 ± 29, rank #7169 of 10916 rated models, from 12 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 30.33 | Accuracy (%) (5-shot) | 47.3 |
| Open LLM Leaderboard v1 - MMLU | 59.52 | Accuracy (%) (5-shot) | 45.3 |
| Open LLM Leaderboard v1 - ARC Challenge | 59.73 | Normalized accuracy (%) (25-shot) | 44.7 |
| Open LLM Leaderboard v1 - HellaSwag | 80.77 | Normalized accuracy (%) (10-shot) | 42.3 |
| Open LLM Leaderboard - MuSR | 39.77 | Score | 40.1 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 48.16 | MC2 (%) (0-shot) | 40 |
| Open LLM Leaderboard v1 - WinoGrande | 74.03 | Accuracy (%) (5-shot) | 34.6 |
| Open LLM Leaderboard - GPQA | 27.94 | Score | 34.2 |
| Open LLM Leaderboard - MMLU-Pro | 27.8 | Score | 30.5 |
| Open LLM Leaderboard - BBH | 43.34 | Score | 30.3 |
| Open LLM Leaderboard - MATH Level 5 | 4.98 | Score | 27.8 |
| Open LLM Leaderboard - IFEval | 23.28 | Score | 16.8 |
Interactive version: theaggregate.ai/model?slug=hebrew-mistral-7b · How It Works · Data refreshed daily, snapshot 2026-11-08.