Mistral Small 3 24B Base 2501
Mistral AI · 2025-01-30 · 23.6B parameters
Mistral Small 3 24B Base 2501 is the pretrained base checkpoint of Mistral AI's latency-optimized 24B Mistral Small 3, released alongside the instruction-tuned checkpoint on January 30, 2025 under Apache 2.0. It is a text-only dense model with a 32,768-token context window and the 131,072-ID Tekken tokenizer.
Benchmark scores
| Benchmark | Score |
|---|---|
| AGIEval English | 65.8 |
| ARC-Challenge | 91.3 |
| Capability | 116.0 |
| GPQA Main | 34.4 |
| GSM8K | 80.7 |
| MATH (Maj1@4) | 46.0 |
| MBPP | 69.6 |
| MMLU | 80.7 |
| MMLU Pro | 54.4 |
| TriviaQA | 80.3 |