DeepSeek V3P2 (Thinking): benchmark results
Provider: DeepSeek. Released 2025-12-01. Access: Open.
Aggregate Rating 1611 ± 28, rank #686 of 2170 rated models, from 13 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Vals AI Poker Agent | 1090.3 | TrueSkill Rating | 80.6 |
| MedQA | 93.92 | Score (%) | 79.8 |
| Vals AI AIME | 84.58 | Accuracy (%) | 62.4 |
| Vals AI MMLU-Pro | 84.92 | Accuracy (%) | 56.9 |
| Vals AI GPQA | 80.3 | Accuracy (%) | 52.3 |
| IOI | 10.67 | Score (imported) | 45 |
| Vals AI CaseLaw v2 | 55.41 | Accuracy (%) | 37.5 |
| Vals AI TaxEval v2 | 68.15 | Accuracy (%) | 29.9 |
| Vals AI LegalBench | 76.08 | Accuracy (%) | 25.3 |
| Vals AI CorpFin v2 | 50.97 | Accuracy (%) | 21.7 |
| Vals AI MGSM | 86.04 | Accuracy (%) | 14.6 |
| ProofBench | 4 | Score (imported) | 10.5 |
Interactive version: theaggregate.ai/model?slug=deepseek-v3p2-thinking · How It Works · Data refreshed daily, snapshot 2026-11-05.