DeepSeek V4 Pro (Reasoning): benchmark results
Provider: DeepSeek. Released 2026-04-24. Access: Open.
Aggregate Rating 1720 ± 17, rank #304 of 2168 rated models, from 29 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| UGI Leaderboard | 62.26 | UGI Score | 99.1 |
| UGI - Writing | 68.4 | Writing Score | 97 |
| UGI - Natural Intelligence | 67.03 | NatInt Score | 95 |
| Wolfram LLM Benchmarking Project | 62.7 | Correct Functionality (%) | 86.8 |
| EntLORE - Cross-Source Composition (Agentic Retrieval) | 39.1 | Answer accuracy (%; L2, 204 questions composing facts stated | 85.7 |
| EntLORE - Cross-Source Composition (LLM Wiki) | 56.9 | Answer accuracy (%; L2, 204 questions composing facts stated | 85.7 |
| EntLORE - Latent Organizational Reasoning (Gold Documents) | 73.5 | Answer accuracy (%; L3, 234 questions whose target relation | 85.7 |
| AtmosCoder-Bench | 93.3 | Accuracy (%, mean of 3 runs) | 80 |
| GRIPS | 92.9 | Accuracy (%) | 76.9 |
| GRIPS-hard | 67.3 | Accuracy (%) | 76.9 |
| Blind-Spots-Bench - Text-only | 70.4 | Accuracy (%; mean@4 over four samples on the text-only quest | 73.3 |
| EntLORE - Cross-Source Composition (GraphRAG) | 48.8 | Answer accuracy (%; L2, 204 questions composing facts stated | 71.4 |
Interactive version: theaggregate.ai/model?slug=deepseek-v4-pro-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.