Raiden-16x3.43B: benchmark results
Kquant03's mixture-of-experts release whose published Mixtral configuration specifies 16 local experts, 2 selected per token and 26 decoder layers. Its recipe names heegyu/WizardVicuna-Uncensored-3B-0719 as the base; nominal expert sizes are not measured total or active parameter counts. All sixteen expert slots cite heegyu/WizardVicuna-Uncensored-3B-0719; repeated slots do not establish sixteen independently trained source checkpoints. Provider: Kquant03. Access: Open.
Aggregate Rating 1178 ± 39, rank #9727 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 2.43 | Accuracy (%) (5-shot) | 20.1 |
| Open LLM Leaderboard v1 - ARC Challenge | 41.89 | Normalized accuracy (%) (25-shot) | 18.1 |
| Open LLM Leaderboard v1 - HellaSwag | 66.2 | Normalized accuracy (%) (10-shot) | 17.6 |
| Open LLM Leaderboard v1 - WinoGrande | 63.61 | Accuracy (%) (5-shot) | 16.4 |
| Open LLM Leaderboard v1 - MMLU | 26.24 | Accuracy (%) (5-shot) | 11.8 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 39.18 | MC2 (%) (0-shot) | 10.3 |
Interactive version: theaggregate.ai/model?slug=raiden-16x3-43b · How It Works · Data refreshed daily, snapshot 2026-11-08.