Magistral Small 1.2

Name: Magistral Small 1.2
Author: Mistral AI

Mistral AI · Magistral · released 2025-09-17 · data checked 2026-11-01 Needs verification

Magistral Small 1.2 is a 24B-parameter dense model from Mistral AI (Europe), released 2025-09-17 under the Apache-2.0 license. It scores 55.9 on the Small LM Index, #23 of 34 medium models. It needs about 16.9 GB of memory at 4-bit with an 8K-token context.

Small LM Index

55.9

#23 of 34 scored medium models

Likely 51.2–60.3 · from 3 of 6 core benchmarks

Parameters

24B

Dense

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

128K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best medium model on that benchmark. Sources: 3 reported by Mistral AI. How we label sources.

GPQA Diamond Lab-reported Knowledge70.1

Class best: Nanbeige4.3 33B 89.5

AIME Lab-reported Math · AIME 202577.3

Class best: Ternary Bonsai 2 27B 95.8

LiveCodeBench Lab-reported Coding · LiveCodeBench v570.9

Class best: Qwen3.8 27B 90.3

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

MMLU-Pro≈ 76

Humanity's Last Exam≈ 9

SWE-bench Verified≈ 39

Specifications

Lab: Mistral AI (FR · Europe)
Architecture: Dense transformer
Total parameters: 24B
Active parameters: All (dense)
Size class: Medium
Context window: 128K tokens
Input modalities: text, image
Reasoning: Reasoning (thinking)
License: Apache-2.0
Openness: Open weights
Memory to run: ~16.9 GB at 4-bit, 8K context (more sizes)
Hugging Face: 29.7K downloads (30 days) · 316 likes

Weights on Hugging Face ↗

Something wrong here? Report it.

Run it locally

Uses mistralai/Magistral-Small-2509-GGUF (Q4_K_M). Swap the tag for another size: BF16, Q4_K_M, Q5_K_M, Q8_0.

Ollama

ollama run hf.co/mistralai/Magistral-Small-2509-GGUF:Q4_K_M

llama.cpp

llama-server -hf mistralai/Magistral-Small-2509-GGUF:Q4_K_M

Then open http://localhost:8080 for a chat window.

MLX (Mac)

pip install -U mlx-lm
mlx_lm.chat --model lmstudio-community/Magistral-Small-2509-MLX-4bit

Text only; for images use mlx-vlm.

vLLM (GPU server)

vllm serve mistralai/Magistral-Small-2509

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	128K context
16-bit F16	50 GB	54 GB	70 GB
8-bit Q8_0	27 GB	32 GB	47 GB
4-bit Q4_K_M	16 GB	21 GB	37 GB

Check it against your GPU, Mac or phone →

Magistral Small 1.2

Benchmarks

Not reported: estimated

Specifications

Compare with

Ranked in

Sources

Run it locally