DeepSeek V3.2 (Non-reasoning): benchmark results

DeepSeek V3.2 evaluated with reasoning disabled. Provider: DeepSeek. Released 2025-12-01. Access: Open.

Unified ELO 1575 ± 1, rank #534 of 1761 rated models, from 61 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
CritPt90Accuracy (self-reported)97.7
UGI Leaderboard55.75UGI Score96.4
UGI - Writing56.3Writing Score91
UGI - Natural Intelligence47.85NatInt Score89.5
AA Terminal-Bench Hard32.58Accuracy (%)76.6
UGI - Willingness (W/10)7.2W/10 Score73.9
AA TAU-2 Bench78.95Accuracy (%)69.1
AA Omniscience - Humanities & Social Sciences25.9Accuracy (%)66.3
AA Omniscience - Law17.3Accuracy (%)65.1
AA Omniscience - Software Engineering (SWE)33.7Accuracy (%)64.5
AA Global-MMLU-Lite - English91.08Accuracy (%)64
Artificial Analysis Intelligence Index18.27Intelligence Index63.1

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