DeepSeek V2 Chat: benchmark results
DeepSeek V2 Chat is a mixture-of-experts model with 21B active parameters per token, as reported by its publisher. Its published checkpoint context is 131,072 tokens; endpoint limits may differ. Provider: DeepSeek. Released 2024-05-06. Access: Open.
Aggregate Rating 1565 ± 16, rank #563 of 1656 rated models, from 127 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenCompass Math - Primary School - English (CompassBench 2404) | 87.2 | Score (%) | 100 |
| OpenCompass Math - Primary School - Chinese (CompassBench 2405) | 92 | Score (%) | 98.7 |
| Open Portuguese LLM - ASSIN2 STS | 85.33 | Pearson Correlation (×100) | 97.6 |
| OpenCompass Code - LCBench - Chinese (CompassBench 2405) | 58.1 | Score (%) | 97.4 |
| OpenCompass LLM - Math - English (CompassBench 2405) | 66.5 | Score (%) | 97.4 |
| OpenCompass Math - Middle School - English (CompassBench 2405) | 57.9 | Score (%) | 97.4 |
| OpenCompass Code - MBPP - Chinese (CompassBench 2405) | 69 | Score (%) | 97.3 |
| Open Portuguese LLM - ASSIN2 RTE | 94.4 | Macro F1 (%) | 96.2 |
| OpenCompass LLM - Code - Chinese (CompassBench 2404) | 67.7 | Score (%) | 96 |
| OpenCompass Math - Primary School - Chinese (CompassBench 2404) | 88.3 | Score (%) | 96 |
| OpenCompass Subjective - Creation (CompassBench 2404) | 54.6 | Score (%) | 96 |
| OpenCompass Code - LCBench - English (CompassBench 2405) | 53.7 | Score (%) | 94.7 |
Interactive version: theaggregate.ai/model?slug=deepseek-v2-chat · How It Works · Data refreshed daily, snapshot 2026-10-20.