DeepSeek V2.5: benchmark results
DeepSeek's open 236B MoE (21B active, 128K context) merging DeepSeek-V2-Chat and Coder-V2 into one general+coding model (September 2024). Provider: DeepSeek. Released 2024-09-05. Access: Open.
Aggregate Rating 1586 ± 11, rank #489 of 1656 rated models, from 178 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenCompass Code - Code QA - Chinese (CompassBench 2409) | 64.7 | Score (%) | 100 |
| OpenCompass Code - Code QA - English (CompassBench 2409) | 65.3 | Score (%) | 100 |
| OpenCompass Code - LCBench - English (CompassBench 2409) | 76.9 | Score (%) | 100 |
| OpenCompass Code - LCBench - Chinese (CompassBench 2409) | 77.8 | Score (%) | 96.6 |
| OpenCompass LLM - Code (CompassBench 2409) | 72.7 | Score (%) | 96.6 |
| OpenCompass LLM - Code - Chinese (CompassBench 2409) | 73.9 | Score (%) | 96.6 |
| OpenCompass LLM - Code - English (CompassBench 2409) | 71.4 | Score (%) | 96.6 |
| OpenCompass Language - Dialogue - Chinese (CompassBench 2409) | 63.5 | Score (%) | 96.6 |
| OpenCompass Agent - T-Eval - English (CompassBench 2409) | 88.1 | Score (%) | 94.8 |
| OpenCompass Language - Creation - Chinese (CompassBench 2409) | 53.4 | Score (%) | 93.1 |
| OpenCompass Language - Dialogue - English (CompassBench 2409) | 54.5 | Score (%) | 93.1 |
| OpenCompass Reasoning - Science and Engineering - Chinese (CompassBench 2409) | 60.6 | Score (%) | 93.1 |
Interactive version: theaggregate.ai/model?slug=deepseek-v2-5 · How It Works · Data refreshed daily, snapshot 2026-10-20.