DeepSeek V3.2 Speciale — benchmark results

DeepSeek V3.2 Speciale variant row. Provider: DeepSeek. Released 2025-12-01. Access: API.

Unified ELO 1765 ± 22, rank #172 of 1776 rated models, from 127 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
MathArena - CMIMC 202594.38Accuracy (%)100
OpenCompass Research - AIME 202596Score (%)100
UGI Leaderboard65.37UGI Score99.5
AA LiveCodeBench89.63Pass@1 (%)99.4
AA AIME 202596.67Accuracy (%)98.3
JudgeBench Reasoning96.94Accuracy (%)98
OpenCompass Research - HLE28.6Score (%)96.8
AA MMLU-Pro86.31Accuracy (%)95.3
BenchTable76.9Total Score (%)94.5
LLM Stats (HMMT 2025)99.2Score (%)93.8
MathArena - BRUMO 202599.17Accuracy (%)92.2
CritPt7.4Accuracy (self-reported)92

Interactive version: theaggregate.ai/model?slug=deepseek-v3-2-speciale · How the rankings work · Data refreshed daily, snapshot 2026-07-22.