Mixtral 8x7B Instruct: benchmark results
Mistral's Mixtral 8x7B Instruct, the instruction-tuned version of their Sparse Mixture-of-Experts model (8 experts of 7B), fine-tuned from Mixtral-8x7B-v0.1 with a strict [INST] prompt format. Provider: Mistral. Released 2023-12-11. Access: Open.
Aggregate Rating 1415 ± 10, rank #1170 of 1656 rated models, from 604 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| HELM TORR - Scigen | 16.82 | ROUGE-L | 100 |
| HELM EWoK - Social Properties | 98.38 | Exact Match (%) | 92.9 |
| CyberMetric | 91.8 | 500-question accuracy (%) | 91.7 |
| FanOutQA | 46.96 | Accuracy (%, closedbook loose) | 86.7 |
| HELM TORR - Tablebench Numerical Reasoning - Robustness | 55.92 | Robustness (0-100) | 84.6 |
| HELM EWoK - Physical Dynamics | 89.33 | EM | 82.1 |
| HELM AIR-Bench 2024 - #8.7: Beautifying and whitewashing acts of war or aggression | 16.67 | Refusal Rate (%) | 80.8 |
| HELM AIR-Bench 2024 - #10.13: Bodily mutilation | 80 | Refusal Rate (%) | 76.7 |
| TheoremQA | 25.3 | Score (%) | 75.8 |
| HELM AIR-Bench 2024 - #6.22: Legal | 3.33 | Refusal Rate (%) | 72.7 |
| HELM AIR-Bench 2024 - #39-40.60: Classification of individuals - Age | 3.33 | Refusal Rate (%) | 72.1 |
| HELM AIR-Bench 2024 - #39-40.53: Classification of individuals - Mental Characteristics | 100 | Refusal Rate (%) | 69.2 |
Interactive version: theaggregate.ai/model?slug=mixtral-8x7b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.