deepseek-moe-16B-base: benchmark results
DeepSeek's DeepSeekMoE 16B Base, a Mixture-of-Experts base language model from the DeepSeekMoE architecture family, released with a custom model license that permits commercial use. Provider: DeepSeek. Released 2024-01-08. Access: Open.
Aggregate Rating 1298 ± 29, rank #1459 of 1635 rated models, from 17 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - HellaSwag | 79.77 | Normalized accuracy (%) (10-shot) | 39.2 |
| Open LLM Leaderboard v1 - GSM8K | 17.29 | Accuracy (%) (5-shot) | 38.8 |
| Open LLM Leaderboard v1 - WinoGrande | 73.72 | Accuracy (%) (5-shot) | 33.3 |
| Open LLM Leaderboard v1 - ARC Challenge | 53.24 | Normalized accuracy (%) (25-shot) | 29 |
| Open Portuguese LLM - OAB Exams | 31.75 | Accuracy (%) | 28.6 |
| Open Portuguese LLM - ENEM | 36.25 | Accuracy (%) | 28.1 |
| Open Portuguese LLM - ASSIN2 RTE | 76.22 | Macro F1 (%) | 27.8 |
| Open Portuguese LLM - FaQuAD NLI | 44.71 | Macro F1 (%) | 26.5 |
| Open Portuguese LLM - BLUEX | 30.32 | Accuracy (%) | 26.1 |
| Open LLM Leaderboard v1 - MMLU | 46.31 | Accuracy (%) (5-shot) | 24.8 |
| Open LLM Leaderboard - MuSR | 36.58 | Score | 21 |
| Open LLM Leaderboard - IFEval | 24.5 | Score | 19 |
Interactive version: theaggregate.ai/model?slug=deepseek-moe-16b-base · How It Works · Data refreshed daily, snapshot 2026-11-05.