Atlas

Models

← All models

Mistral Small Instruct 2409

Mistral AI · 2024-09-17 · 22.2B parameters

Mistral Small Instruct 2409 is the instruction-fine-tuned September 2024 checkpoint from Mistral AI, a 22-billion-parameter dense text model with a 32K sequence length and function-calling support. Released as an upgrade to Mistral Small v24.02 under the Mistral Research License, it was positioned for cost-efficient tasks such as translation, summarization, and sentiment analysis, with official release materials reporting improvements in human alignment, reasoning, and code.

Benchmark scores

BenchmarkScore
Artificial Analysis Intelligence Index4.7
BBH (Open LLM Leaderboard v2)52.1
BigCodeBench Complete46.6
BigCodeBench Instruct36.1
BigCodeBench-Hard Complete18.9
BigCodeBench-Hard Instruct10.8
BRIDGE Medical (chain-of-thought)31.2
BRIDGE Medical (few-shot)36.0
BRIDGE Medical (zero-shot)35.2
Capability113.5
Creative Writing v272.4
EQ-Bench v280.9
EQ-Bench v2 + MAGI-Hard Combined66.2
Global PIQA - Non-Parallel (Strict Exact Match)54.5
Global PIQA - Parallel (Strict Exact Match)26.7
GPQA (Open LLM Leaderboard v2)32.4
GPQA Diamond38.1
Humanity's Last Exam (Text-Only)4.3
IFEval (Open LLM Leaderboard v2)66.7
LiveCodeBench (Artificial Analysis)14.1
MAGI-Hard51.6
MATH Level 5 (Open LLM Leaderboard v2)14.4
MATH-500 (Artificial Analysis source)56.3
MMLU Pro52.9
MMLU-Pro (Open LLM Leaderboard v2)39.6
MuSR (Open LLM Leaderboard v2)36.3
SciCode (Artificial Analysis)15.6
Loading Atlas data…