Model

Magistral Small

OPEN

Mistral AI · Magistral family · released Jun 10, 2025

Mistral's first open-weights reasoning model (Apache 2.0); transparent multilingual chain-of-thought, fits a single RTX 4090 or 32GB Mac once quantized.

ReasoningCodingVisionFunction callingTool useAgentic

1422.4

Elo · rank #134

Parameters

24B

Active params

24B (dense)

Context

128K tokens

Architecture

Dense transformer (24B), reasoning fine-tune of Mistral Small 3.1 with cold-start data distilled from Magistral Medium

License

Apache 2.0

Languages

24+

API price (in/out)

$0.5 / $1.5

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

AIMEMath70.7%#102

best: GPT-5.2 · 100.0%

GPQA DiamondReasoning68.2%#130

best: GPT-5.6 · 94.6%

Humanity's Last ExamReasoning6.4%#109

best: Claude Sonnet 5 · 57.4%

LiveCodeBenchCoding55.8%#109

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MATH-500Math95.9%#30

best: GPT-5 · 99.4%

Run it locally

VRAM @ Q4

14.5 GB

VRAM @ FP16

55 GB

Fits on (Q4)

RTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Vendor states it fits and runs on a single RTX 4090 once quantized; no independently measured tok/s figure found

Quantizations

GGUF Q4 · MLX

Magistral family

Elo progression across releases

Magistral MediumJun 20251485.1 Magistral SmallJun 20251422.4

Resources

HFmistralai/Magistral-Small-2506 DOCSdocs.mistral.ai APIAvailable via Mistral AI API

API price $0.5/$1.5 · each benchmark row carries its own source badge (see methodology)