Atlas

Models

← All models

Mistral Small 1.0

Mistral AI · 2024-02-26

Mistral Small 1.0 (`mistral-small-2402`) was Mistral AI's February 2024 enterprise-grade text model optimized for latency, cost, and efficiency. Released alongside Mistral Large, it supported RAG, function calling, and JSON mode; Mistral published only a rounded 32k context and no exact parameter count. Mistral retired it on June 16, 2025 and lists Mistral Small 4 as its replacement.

Benchmark scores

BenchmarkScore
AIME 2024 (avg@8)7.9
AIME 2024-20255.6
AIME 2025 (avg@8)3.3
Artificial Analysis Intelligence Index3.6
BigCodeBench Complete41.3
BigCodeBench Instruct32.1
BigCodeBench-Hard Complete14.9
BigCodeBench-Hard Instruct12.8
Capability112.6
EQ-Bench v280.4
EQ-Bench v2 + MAGI-Hard Combined66.1
GPQA Diamond44.9
GPQA Diamond (Vals protocol average)50.8
Humanity's Last Exam (Text-Only)4.4
LiveCodeBench (Artificial Analysis)11.1
LiveCodeBench (Vals AI public dataset)15.8
LiveCodeBench (Vals Public) - Easy42.9
LiveCodeBench (Vals Public) - Hard0.6
LiveCodeBench (Vals Public) - Medium3.9
MAGI-Hard51.9
MATH-50070.6
MATH-500 (Artificial Analysis source)56.2
MedQA Demographic Bias (Vals)57.0
MedQA Demographic Bias (Vals) - Asian56.6
MedQA Demographic Bias (Vals) - Black56.6
MedQA Demographic Bias (Vals) - Hispanic56.5
MedQA Demographic Bias (Vals) - Indigenous56.9
MedQA Demographic Bias (Vals) - Unbiased58.0
MedQA Demographic Bias (Vals) - White57.3
MGSM84.0
Loading Atlas data…