Mixtral 8x22B Instruct — benchmark results

Mistral's Apache-2.0 sparse-MoE instruct model (141B total/39B active, 64K context), strong at math and coding among April 2024 open models. Provider: Mistral. Released 2024-04-17. Access: Open.

Unified ELO 1559 ± 14, rank #596 of 1776 rated models, from 95 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AGC-Bench - grapheval_ai_researcher0.81Dataset z-score87.7
AGC-Bench - tinystories0.94Dataset z-score85.4
AGC-Bench - story_quality0.98Dataset z-score84.6
AGC-Bench - future_ideas0.88Dataset z-score79.3
AGC-Bench - schnovel0.9Dataset z-score78.7
AGC-Bench - thenextchapter0.59Dataset z-score76.8
AGC-Bench - grapheval_iclr0.65Dataset z-score75.6
AGC-Bench - pollux_creativity0.75Dataset z-score75.6
AGC-Bench - grapheval_review_advisor0.52Dataset z-score73.2
AGC-Bench - conceptual_design0.61Dataset z-score70.7
AGC-Bench - metaphoric_analogies0.32Dataset z-score70.7
VNTL Leaderboard68.46Accuracy (%)67.4

Interactive version: theaggregate.ai/model?slug=mixtral-8x22b-instruct · How the rankings work · Data refreshed daily, snapshot 2026-07-22.