DeepSeek V3P2: benchmark results

Provider: DeepSeek. Released 2025-12-01. Access: Open.

Unified ELO 1800 ± 28, rank #155 of 2656 rated models, from 10 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
Vals AI MGSM90.87Accuracy (%)56.9
IOI14.42Score (self-reported)51.7
MMLU Pro83.06Score (%)49.2
GPQA Diamond76.26Score (%)46
MedQA89.45Score (%)45.7
AIME64.79Score (%)43.2
ProofBench8Score (self-reported)19.3
CorpFin v247.94Score (%)17.6
LegalBench65.96Score (%)10.2
Vals AI Finance Agent0Accuracy (%)0.8

Interactive version: theaggregate.ai/model?slug=deepseek-v3p2 · How It Works · Data refreshed daily, snapshot 2026-09-19.