DeepSeek V4 Pro (Max): benchmark results
DeepSeek V4 Pro evaluated at the max reasoning-effort setting. Provider: DeepSeek. Released 2026-04-24. Access: Open.
Aggregate Rating 1752 ± 7, rank #234 of 2168 rated models, from 274 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AdvancedMathBench - VerifierBench | 65.1 | Meta-verification balanced F1 (%): a validity verdict counts | 100 |
| LG K-EXAONE 2.0 Card - CLIcK | 91.6 | Accuracy (%) | 100 |
| LG K-EXAONE 2.0 Card - HRM8K-KSM | 94.3 | Accuracy (%) | 100 |
| LG K-EXAONE 2.0 Card - KMMLU-Pro | 80.5 | Accuracy (%) | 100 |
| LG K-EXAONE 2.0 Card - Ko-LongBench | 91.4 | Score (%) | 100 |
| LLM Stats (CSimpleQA) | 84.4 | Score (%) | 100 |
| LLM Stats (Codeforces Rating - DeepSeek V4 family) | 3206 | Codeforces Rating | 100 |
| MAXIFE | 88.9 | Score (%) | 100 |
| SuperCLUE General (July 2026) - Precise Instruction Following | 49.52 | Score | 100 |
| AGI-Eval Community - Algorithmic Reasoning | 71.92 | Accuracy (%) | 95.7 |
| Nejumi 4 - jaster (2-shot) - NIILC | 60.83 | Character F1 (x100) | 95.1 |
| Nejumi 4 - jaster (0-shot) - AIO | 94.16 | Character F1 (x100) | 94.4 |
Interactive version: theaggregate.ai/model?slug=deepseek-v4-pro-max · How It Works · Data refreshed daily, snapshot 2026-10-20.