Mixtral 8x22B

by Mistral AI · MIXTRAL family · released Apr 17, 2024
Open weights Active

Big open MoE, strong on code and math.

This model has no first-party API. Open-weight release: self-host from published weights.

Specifications

Model typeText LLM
ArchitectureMoE
Parameters141 (39 active)
Context window65.5K tokens
Max outputNot disclosed tokens
Knowledge cutoffNot disclosed
LicenseApache License 2.0
Modalities Text input Text output
Tags Long context
End of lifeNot disclosed
Last verifiedAug 20, 2026

Benchmarks

BenchmarkScoreSrc
MMLU 77.8% official

Family versions