Ministral 3 14B Reasoning 2512
Mistral AI · 2025-12-02 · 13.9B parameters
Ministral 3 14B Reasoning 2512 is the reasoning post-trained checkpoint of Mistral AI's December 2025 Ministral 3 14B, released alongside the separate Base and non-reasoning Instruct checkpoints. The open-weight Apache-2.0 model accepts text and image input, produces text with always-on reasoning traces for math, coding and STEM tasks, and supports a 262,144-token context.
Benchmark scores
| Benchmark | Score |
|---|---|
| AIME 2024 | 89.8 |
| AIME 2025 | 85.0 |
| GPQA Diamond | 71.2 |
| LiveCodeBench v6 - release-reported cohort | 64.6 |