UNAversal-8x7B-v1beta: benchmark results
fblgit experimental UNA-SFT derivative of Mixtral 8x7B Instruct, with eight MLP experts per decoder layer. Provider: fblgit. Access: Open.
Aggregate Rating 1469 ± 34, rank #2862 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - MMLU | 70.39 | Accuracy (%) (5-shot) | 93.9 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 71.97 | MC2 (%) (0-shot) | 92.8 |
| Open LLM Leaderboard v1 - ARC Challenge | 69.8 | Normalized accuracy (%) (25-shot) | 84.3 |
| Open LLM Leaderboard v1 - HellaSwag | 86.9 | Normalized accuracy (%) (10-shot) | 83.6 |
| Open LLM Leaderboard v1 - WinoGrande | 82 | Accuracy (%) (5-shot) | 81.7 |
| Open LLM Leaderboard v1 - GSM8K | 61.64 | Accuracy (%) (5-shot) | 74.4 |
Interactive version: theaggregate.ai/model?slug=unaversal-8x7b-v1beta · How It Works · Data refreshed daily, snapshot 2026-11-08.