DeepSeek V3.2 Exp (Thinking): benchmark results
DeepSeek V3.2 Exp evaluated with thinking enabled. Provider: DeepSeek. Released 2025-09-29. Access: Open.
Aggregate Rating 1641 ± 10, rank #554 of 2168 rated models, from 176 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| UGI Leaderboard | 53.84 | UGI Score | 95.1 |
| LogicGraph - Precision | 76.96 | Precision (%): share of the model's generated proof paths th | 93.3 |
| AA Omniscience - Software Engineering (SWE) - Swift | 56 | Accuracy (%) | 91.5 |
| AA LiveCodeBench | 78.94 | Pass@1 (%) | 91.1 |
| AA MMLU-Pro | 85.04 | Accuracy (%) | 90.9 |
| UGI - Writing | 54 | Writing Score | 88.8 |
| UGI - Natural Intelligence | 46.34 | NatInt Score | 87.1 |
| LogicGraph | 88 | Success rate (%): share of problems with at least one fully | 86.7 |
| AA AIME 2025 | 87.67 | Accuracy (%) | 85.1 |
| Chatbot Arena (Text - Medicine & Healthcare) | 1475 | Arena Score | 83.1 |
| AA Global-MMLU-Lite - German | 91.08 | Accuracy (%) | 82.7 |
| AA Global-MMLU-Lite - French | 91.08 | Accuracy (%) | 81.8 |
Interactive version: theaggregate.ai/model?slug=deepseek-v3-2-exp-thinking · How It Works · Data refreshed daily, snapshot 2026-10-20.