DeepSeek V3.2 (Thinking): benchmark results

DeepSeek V3.2 evaluated with thinking enabled. Provider: DeepSeek. Released 2025-12-01. Access: Open.

Unified ELO 1617 ± 1, rank #362 of 1761 rated models, from 136 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
SuperCLUE-Mkt - Compliance Assessment92.34Score100
SuperCLUE-Mkt - Placement Strategy89.66Score100
UGI Leaderboard57.78UGI Score98.1
UGI - Writing57.28Writing Score91.6
SuperCLUE-Mkt - Market Insight88.12Score90.9
SuperCLUE-Mkt - Overall85.74Score90.9
UGI - Natural Intelligence48.11NatInt Score89.7
C4STYLI74.5Accuracy (C4Styli-T) (self-reported)88.2
LLM2014 Logic 2025-1054.16Median Score87.8
BenchTable70.7Total Score (%)87.5
AA TAU-2 Bench90.64Accuracy (%)85.4
LLM2014 Code 2025-11 - TypeScript8.17Score84

Interactive version: theaggregate.ai/model?slug=deepseek-v3-2-thinking · How It Works · Data refreshed daily, snapshot 2026-09-05.