Nanbeige2-16B-Chat: benchmark results
Nanbeige Lab's 16B chat model trained on 4.5T tokens and aligned with 1M-sample SFT, curriculum learning, and DPO (May 2024). Provider: Nanbeige. Released 2024-05-24.
Aggregate Rating 1453 ± 17, rank #978 of 1656 rated models, from 156 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenCompass Agent - T-Eval - Chinese (CompassBench 2405) | 90.4 | Score (%) | 100 |
| OpenCompass Agent - T-Eval - Chinese (CompassBench 2407) | 90.3 | Score (%) | 100 |
| OpenCompass LLM - Agent - Chinese (CompassBench 2405) | 76.4 | Score (%) | 92.1 |
| OpenCompass LLM - Language - Chinese (CompassBench 2407) | 54.4 | Score (%) | 91.2 |
| OpenCompass Language - Creation - Chinese (CompassBench 2407) | 50 | Score (%) | 91.2 |
| OpenCompass Language - Dialogue - Chinese (CompassBench 2407) | 58.2 | Score (%) | 91.2 |
| OpenCompass Subjective - Knowledge - English (CompassBench 2405) | 52.1 | Score (%) | 90.8 |
| OpenCompass Subjective - Knowledge (CompassBench 2405) | 54.7 | Score (%) | 89.5 |
| OpenCompass Subjective - Knowledge - Chinese (CompassBench 2405) | 57.3 | Score (%) | 89.5 |
| Open Chinese LLM - ARC Challenge | 60.32 | Accuracy (%) | 88.2 |
| OpenCompass LLM - Agent (CompassBench 2407) | 85.8 | Score (%) | 88.2 |
| Open LLM Leaderboard v1 - MMLU | 66.63 | Accuracy (%) (5-shot) | 87.8 |
Interactive version: theaggregate.ai/model?slug=nanbeige2-16b-chat · How It Works · Data refreshed daily, snapshot 2026-10-20.