DeepSeek V4 Flash (Reasoning): benchmark results
Provider: DeepSeek. Released 2026-04-24. Access: Open.
Aggregate Rating 1710 ± 15, rank #329 of 2168 rated models, from 32 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EntLORE - Cross-Source Composition (GraphRAG) | 51.7 | Answer accuracy (%; L2, 204 questions composing facts stated | 100 |
| EntLORE - Cross-Source Composition (LLM Wiki) | 59.6 | Answer accuracy (%; L2, 204 questions composing facts stated | 100 |
| SKT A.X K2 Card - KMO 2026 (1st Round) | 94.4 | Accuracy (%) | 100 |
| SKT A.X K2 Card - KoBALT | 75.3 | Accuracy (%) | 100 |
| UGI Leaderboard | 56.63 | UGI Score | 97.3 |
| UGI - Natural Intelligence | 45.52 | NatInt Score | 86.8 |
| UGI - Writing | 49.74 | Writing Score | 86.5 |
| EntLORE - Explicit Lookup (GraphRAG) | 67.6 | Answer accuracy (%; L1, 469 questions answered by a stated f | 85.7 |
| EntLORE - Latent Organizational Reasoning (Agentic Retrieval) | 15.1 | Answer accuracy (%; L3, 234 questions whose target relation | 85.7 |
| EntLORE - Latent Organizational Reasoning (GraphRAG) | 39.8 | Answer accuracy (%; L3, 234 questions whose target relation | 85.7 |
| SKT A.X K2 Card - CLIcK | 89.6 | Accuracy (%) | 85.7 |
| SKT A.X K2 Card - KMMLU-Pro | 78.8 | Accuracy (%) | 85.7 |
Interactive version: theaggregate.ai/model?slug=deepseek-v4-flash-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.