Ministral 3 14B: benchmark results
Mistral's Apache-2.0 14B dense model from the Mistral 3 family, with image understanding and single-GPU focus (December 2025). Provider: Mistral. Released 2025-12-02. Access: Open.
Aggregate Rating 1538 ± 7, rank #650 of 1656 rated models, from 490 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Konkur 1404 - Foreign Language | 44.23 | Accuracy (%, text-only) | 100 |
| SpeechMap Compliance | 93.3 | % Requests Completed | 96.1 |
| MMOOC - Partial Answerability (Yes/No) | 81 | Shifted in-context score (%; mean of accuracy on the answera | 88.2 |
| MMOOC - Unclear Logical and Symbolic (Multiple Choice) | 42 | Out-of-context score (%; mean of the refusal rate, the share | 88.2 |
| 6G-Bench | 79.5 | Accuracy (%) on the full set of 6G-Bench's 3,722 expert-vali | 85.7 |
| 6G-Bench - Distributed Intelligence and Emerging Use Cases | 77 | Accuracy (%) on the distributed intelligence and emerging 6G | 85.2 |
| MMOOC - Multimodal Ambiguity (Multiple Choice) | 50.25 | Out-of-context score (%; mean of the refusal rate, the share | 82.4 |
| MMOOC - Uncertain Spatial and Physical Context (Multiple Choice) | 75.5 | Out-of-context score (%; mean of the refusal rate, the share | 82.4 |
| MMOOC - Uncertain Spatial and Physical Context (Yes/No) | 28.5 | Out-of-context score (%; mean of the refusal rate, the share | 82.4 |
| MMOOC - Unclear Logical and Symbolic (Yes/No) | 45.75 | Out-of-context score (%; mean of the refusal rate, the share | 82.4 |
| PanoSpace-Bench - Relative 3D Position | 31.6 | Accuracy (%; 250 four-option multiple-choice questions on on | 81.8 |
| 6G-Bench - Intent and Policy Reasoning | 87 | Accuracy (%) on the intent and policy reasoning tasks (T1, T | 81.5 |
Interactive version: theaggregate.ai/model?slug=ministral-3-14b · How It Works · Data refreshed daily, snapshot 2026-10-20.