24B multimodal instruction model by Mistral AI, tuned for accuracy, tool use & fewer repeats
10K+
GGUF version by Unsloth
A 24B-parameter instruction-tuned multimodal model by Mistral AI, optimized for following instructions, reducing repetition errors, and supporting function/tool calling.
| Attribute | Details |
|---|---|
| Provider | Mistral AI |
| Architecture | Llama |
| Cutoff date | 2023-10-01 |
| Languages | 24 languages |
| Tool calling | ✅ |
| Input modalities | Text, Images |
| Output modalities | Text |
| License | Apache 2.0 |
| Model variant | Parameters | Quantization | Context window | VRAM¹ | Size |
|---|---|---|---|---|---|
ai/magistral-small-3.2:24Bai/magistral-small-3.2:24B-UD-Q4_K_XLai/magistral-small-3.2:latest | 23.57 B | MOSTLY_Q4_K_M | 41K tokens | 14.59 GiB | 13.50 GB |
ai/magistral-small-3.2:24B-UD-IQ2_XXS | 23.57 B | MOSTLY_IQ2_XXS | 41K tokens | 7.45 GiB | 6.28 GB |
¹: VRAM estimated based on model characteristics.
latest→24B
docker model run ai/magistral-small-2506
| Category | Metric | Mistral-Small-3.2-24B-Instruct-2506 |
|---|---|---|
| Instruction Following | WildBench v2 | 65.33% |
| Arena Hard v2 | 43.1% | |
| IF Accuracy (Internal) | 84.78% | |
| STEM | MMLU | 80.50% |
| MMLU Pro (5-shot CoT) | 69.06% | |
| MATH | 69.42% | |
| GPQA Diamond (5-shot CoT) | 46.13% | |
| MBPP Plus Pass@5 | 78.33% | |
| HumanEval Plus Pass@5 | 92.90% | |
| SimpleQA (TotalAcc) | 12.10% | |
| Vision | MMMU | 62.5% |
| ChartQA | 87.4% | |
| DocVQA | 94.86% | |
| AI2D | 92.91% |
Content type
Model
Digest
sha256:9c3bc0d23…
Size
14.3 GB
Last updated
10 months ago
docker model pull ai/magistral-small-3.2:24B-UD-Q4_K_XLPulls:
108
Last week