Dolphin3.0-R1-Mistral-24B: benchmark results
Cognitive Computations' Dolphin 3.0 R1, a 24B instruct-tuned reasoning model built on Mistral-Small-24B-Base-2501 and trained on 800,000 reasoning traces from the Dolphin-R1 dataset. Provider: Cognitive Computations. Released 2025-02-06. Access: Open.
Aggregate Rating 1367 ± 25, rank #1342 of 1656 rated models, from 184 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard - MATH Level 5 | 31.19 | Score | 82.6 |
| Open LLM Leaderboard - BBH | 53.6 | Score | 66.8 |
| Open Portuguese LLM - BLUEX | 56.19 | Accuracy (%) | 65.5 |
| Open Portuguese LLM - OAB Exams | 44.05 | Accuracy (%) | 57.4 |
| Open Arabic LLM - Arabic MMLU HT Abstract Algebra | 29 | Accuracy (%) | 54.9 |
| Open Portuguese LLM - ASSIN2 RTE | 90.18 | Macro F1 (%) | 54.8 |
| Open Portuguese LLM - ENEM | 62.7 | Accuracy (%) | 53 |
| Open Portuguese LLM - FaQuAD NLI | 66.76 | Macro F1 (%) | 52.9 |
| Open Arabic LLM - Arabic MMLU LAW (Professional) | 63.38 | Accuracy (%) | 50.9 |
| Open LLM Leaderboard - GPQA | 29.45 | Score | 49.7 |
| Open Japanese LLM - CG | 7.83 | Score (%) | 47.8 |
| Open Japanese LLM - Jamp Exact Match | 58.33 | Score (%) | 43.5 |
Interactive version: theaggregate.ai/model?slug=dolphin3-0-r1-mistral-24b · How It Works · Data refreshed daily, snapshot 2026-10-20.