DeepSeek-V3.2-Thinking-Speciale: benchmark results

Provider: DeepSeek. Access: Open.

Unified ELO 1866 ± 33, rank #90 of 2928 rated models, from 22 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
PM-LLM-Benchmark37.5Score94.6
LisanBench0.33Mean Path Length / Current Maximum91.5
AGI-Eval Community - Algorithmic Reasoning65.72Accuracy (%)87.1
AGI-Eval Community - Learning (English)94.55Accuracy (%)84.1
AGI-Eval Community - Subject Reasoning (Chinese)90.27Accuracy (%)83.7
AGI-Eval Community - Mathematical Reasoning84.46Accuracy (%)83.6
AGI-Eval Community - Objective Accuracy (Chinese)73.73Accuracy (%)83.3
AGI-Eval Community - Subject Reasoning86.77Accuracy (%)82.9
AGI-Eval Community - Objective Accuracy76.78Accuracy (%)81.4
AGI-Eval Community - Subject Reasoning (English)85.21Accuracy (%)81.2
AGI-Eval Community - Interaction (Chinese)82.51Accuracy (%)77.5
AGI-Eval Community - Subject Knowledge86.91Accuracy (%)77.1

Interactive version: theaggregate.ai/model?slug=deepseek-v3-2-thinking-speciale · How It Works · Data refreshed daily, snapshot 2026-09-23.