Ryu-4x7B-MoE-bf16: benchmark results
Kquant03's mixture-of-experts release whose published Mixtral configuration specifies 4 local experts, 2 selected per token and 32 decoder layers. Its recipe names fblgit/UNA-TheBeagle-7b-v1 as the base; nominal expert sizes are not measured total or active parameter counts. Provider: Kquant03. Access: Open.
Aggregate Rating 1424 ± 34, rank #4962 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 64.96 | MC2 (%) (0-shot) | 82.8 |
| Open LLM Leaderboard v1 - ARC Challenge | 66.47 | Normalized accuracy (%) (25-shot) | 73.7 |
| Open LLM Leaderboard v1 - WinoGrande | 79.24 | Accuracy (%) (5-shot) | 68.1 |
| Open LLM Leaderboard v1 - MMLU | 63.89 | Accuracy (%) (5-shot) | 63.9 |
| Open LLM Leaderboard v1 - GSM8K | 49.73 | Accuracy (%) (5-shot) | 62.3 |
| Open LLM Leaderboard v1 - HellaSwag | 83.1 | Normalized accuracy (%) (10-shot) | 56 |
Interactive version: theaggregate.ai/model?slug=ryu-4x7b-moe-bf16 · How It Works · Data refreshed daily, snapshot 2026-11-08.