DeepSeek V4 Pro (Medium): benchmark results
Provider: DeepSeek. Released 2026-04-24. Access: Open.
Aggregate Rating 1638 ± 33, rank #573 of 2205 rated models, from 15 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| IKP - T1 | 100 | Accuracy (%) | 86.1 |
| IKP - T2 | 99.5 | Accuracy (%) | 72.2 |
| IKP - T3 | 96.72 | Accuracy (%) | 70.9 |
| Deception by Omission | 54.5 | Chat replies that disclose and own a planted mistake of the | 50 |
| ReplaySCM - Hidden-order | 0.38 | HeldoutWorldExact (mean share of held-out intervention world | 50 |
| Thinkingbox | 43.26 | Pass@1 (%, 20 trials per task) | 43.8 |
| BIABench | 52.7 | Outcome score (0-100): mean over 16 end-to-end bioimage-anal | 40 |
| ReplaySCM - Block-order | 0.41 | HeldoutWorldExact (mean share of held-out intervention world | 37.5 |
| ReplaySCM - Ordered | 0.52 | HeldoutWorldExact (mean share of held-out intervention world | 37.5 |
| Deception by Omission - Chat (Reminded) | 70 | Chat replies that disclose and own the planted mistake when | 33.3 |
| Deception by Omission - Agentic | 3.3 | Agent replies that disclose and own a planted harmful action | 16.7 |
| Deception by Omission - Agentic (Reminded) | 27.3 | Agent replies that disclose and own the planted harmful acti | 16.7 |
Interactive version: theaggregate.ai/model?slug=deepseek-v4-pro-medium · How It Works · Data refreshed daily, snapshot 2026-11-13.