Magistral Small 1.2

Name: Magistral Small 1.2
Author: Mistral AI

Mistral AI · 2025-09-18 · 24B parameters

Magistral Small 1.2 is Mistral AI's September 2025 update to the 24B Magistral Small reasoning model, adding multimodal vision capabilities to the existing chain-of-thought reasoning framework. It is built on Mistral Small 3.2 and is compact enough to run on a 32 GB MacBook once quantized.

Benchmark scores

Benchmark	Score
AA-LCR	16.3
AA-LCR v1.1	19.3
AA-Omniscience Index	-65.1
AIME 2024 (avg@8)	84.6
AIME 2024-2025	80.7
AIME 2025 (Artificial Analysis avg@10)	80.3
AIME 2025 (avg@8)	76.8
Artificial Analysis Agentic Index	1.4
Artificial Analysis Coding Index	14.7
Artificial Analysis Intelligence Index	11.3
Artificial Analysis Omniscience Accuracy	13.4
Artificial Analysis Omniscience Hallucination Rate	90.8
Artificial Analysis Openness Index	50.0
Artificial Analysis Progress Index v1	202
AutomationBench-AA 1.0.6	0.2
Capability	129.7
Chess Puzzles	3.0
CorpFin v2	44.0
CorpFin v2 - Exact Pages	55.5
CorpFin v2 - Max Fitting Context	36.6
CorpFin v2 - Shared Max Context	40.0
CritPt (avg@5)	0.3
Design Arena Overall Code Categories	984
DTBench	61.3
GDPval-AA v2	213
GDPval-AA v2 (normalized)	0.0
GDPval-AA v2.1	12.0
Global MMLU Lite	76.3
GPQA Diamond	60.1
GPQA Diamond (Artificial Analysis avg@5)	66.3