DBRX: benchmark results
Databricks' fine-grained MoE model (132B total/36B active, 16 experts pick 4) trained on 12T tokens, released March 2024 under the Databricks Open Model License. Provider: Databricks. Released 2024-03-27.
Aggregate Rating 1440 ± 11, rank #1045 of 1656 rated models, from 205 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - MMLU | 73.72 | Accuracy (%) (5-shot) | 95.7 |
| Open CoT - LSAT Logical Reasoning | 20.59 | CoT Gain (%) | 95 |
| Open LLM Leaderboard v1 - HellaSwag | 88.85 | Normalized accuracy (%) (10-shot) | 94.4 |
| Open CoT - LSAT Reading Comprehension | 21.19 | CoT Gain (%) | 90.1 |
| OpenCompass LLM - Knowledge - English (CompassBench 2405) | 81.9 | Score (%) | 89.5 |
| URIAL-Bench - Math | 5.9 | Judge Score (0-10) | 88.9 |
| URIAL-Bench - Overall | 7.22 | Judge Score (0-10) | 88.9 |
| URIAL-Bench - Roleplay | 7.93 | Judge Score (0-10) | 88.9 |
| URIAL-Bench - Turn 1 | 8.04 | Judge Score (0-10) | 88.9 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 67.02 | MC2 (%) (0-shot) | 85.3 |
| Open LLM Leaderboard - GPQA | 34.14 | Score | 84.9 |
| Open LLM Leaderboard v1 - GSM8K | 67.32 | Accuracy (%) (5-shot) | 83.4 |
Interactive version: theaggregate.ai/model?slug=dbrx · How It Works · Data refreshed daily, snapshot 2026-10-20.