Mixtral 8x7B
by Mistral AI
· MIXTRAL family · released Dec 11, 2023
Open weights
Active
First famous open MoE: 8 experts, 12.9B active.
This model has no first-party API. Open-weight release: self-host from published weights.
Specifications
| Model type | Text LLM |
|---|---|
| Architecture | MoE |
| Parameters | 46.7 (12.9 active) |
| Context window | 32.8K tokens |
| Max output | 8.2K tokens |
| Knowledge cutoff | Not disclosed |
| License | Apache License 2.0 |
| Modalities | Text input Text output |
| Tags | Flagship |
| End of life | Not disclosed |
| Last verified | Aug 20, 2026 |
Benchmarks
| Benchmark | Score | Src |
|---|---|---|
| MMLU | 71.3% | official |
| MATH | 28.4% | official |