Codestral-22B-v0.1 — benchmark results

Mistral's 22B code-generation model (May 2024) covering 80+ programming languages with fill-in-the-middle, under the non-production MNPL license. Provider: Mistral. Released 2024-05-29. Access: Open.

Unified ELO 1416 ± 26, rank #1152 of 1776 rated models, from 14 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
EvalPlus (HumanEval+ & MBPP+)67.8Pass@1 avg (%)75.8
Open LLM Leaderboard - IFEval57.72Score69.7
BigCodeBench41.8Pass@1 (%)64.3
DuckDB-NSQL54.7Execution Accuracy (%)57.3
Open LLM Leaderboard - MuSR10.74Score54.6
Open LLM Leaderboard - GPQA6.49Score54.5
Open LLM Leaderboard - BBH30.74Score54.4
CodeElo385Elo Rating52.9
Open LLM Leaderboard - MATH Level 510.05Score47.6
Open LLM Leaderboard - MMLU-Pro23.95Score42.3
LiveOIBench15.82Avg Human Percentile22.8
IFEval Leaderboard60.74Final Score20

Interactive version: theaggregate.ai/model?slug=codestral-22b-v0-1 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.