DeepSeek
DeepSeek-V3.2 benchmark results
14 results on 8 benchmarks from 5 sources.
Also listed as: deepseek-v3.2 · deepseek-v3.2-thinking · deepseek-chat · deepseek-reasoner · DeepSeek-V3.2 (Thinking) · deepseek/deepseek-v3.2 · DeepSeek-V3.2 (Thinking; Novita) · DeepSeek V3.2 · DeepSeek V3.2 (high) · DeepSeek-V3.2-Thinking
All benchmark results
| Benchmarks | Score | Rank | Setting | Source type | Date |
|---|---|---|---|---|---|
| LMArena Text | 1425 | 122 | Overall, style control on; 48114 votes; 95% CI 1421–1428“deepseek-v3.2” | Official leaderboard | Oct 8, 2026 |
| LMArena Text | 1423 | 126 | Overall, style control on; 42076 votes; 95% CI 1419–1426“deepseek-v3.2-thinking” | Official leaderboard | Oct 8, 2026 |
| LMArena WebDev | 1362 | 104 | Overall; 6921 votes; 95% CI 1353–1371“deepseek-v3.2-thinking” | Official leaderboard | Oct 8, 2026 |
| LMArena WebDev | 1325 | 118 | Overall; 9386 votes; 95% CI 1317–1333“deepseek-v3.2” | Official leaderboard | Oct 8, 2026 |
| GPQA Diamond | 83.4% | 95 | Epoch AI run, accuracy; effort: Thinking | Independent run | Dec 16, 2025 |
| GPQA Diamond | 71.2% | 158 | Epoch AI run, accuracy | Independent run | Jul 16, 2026 |
| MMLU-Pro | 85.0% | 33 | Overall accuracy; data source: Self-Reported; 685B“DeepSeek-V3.2-Thinking” | Vendor-reported | Mar 11, 2026 |
| SWE-bench Verified | 70.0% | 28 | agent: mini-SWE-agent“deepseek-v3.2” | Official leaderboard | Feb 17, 2026 |
| AIME (OTIS Mock AIME 2024–2025) | 87.8% | 80 | Epoch AI run, OTIS Mock AIME 2024–2025; effort: Thinking | Independent run | Dec 16, 2025 |
| AIME (OTIS Mock AIME 2024–2025) | 48.9% | 169 | Epoch AI run, OTIS Mock AIME 2024–2025 | Independent run | Jul 16, 2026 |
| τ²-bench | 63.8% | 13 | airline domain, pass^1; submitted by DeepSeek; vv0.1.3 | Vendor-reported | Dec 1, 2025 |
| τ²-bench | 81.1% | 10 | retail domain, pass^1; submitted by DeepSeek; vv0.1.3 | Vendor-reported | Dec 1, 2025 |
| τ²-bench | 96.2% | 5 | telecom domain, pass^1; submitted by DeepSeek; vv0.1.3 | Vendor-reported | Dec 1, 2025 |
| Terminal-Bench 2.0 | 39.5% | 102 | Terminal-Bench 2.0; agent: Terminus 2; effort: Thinking; Novita | Official leaderboard via Epoch AI | Dec 1, 2025 |
Related on this site
FAQ
Where do these DeepSeek-V3.2 scores come from?
From Epoch AI Benchmarking Hub, LMArena leaderboard dataset, MMLU-Pro leaderboard (TIGER-Lab), SWE-bench leaderboards (swebench.com), and τ²-bench leaderboard submissions. Each row links to the original leaderboard, dataset or report, and vendor-reported numbers are labelled.