DeepSeek-V3.2-Exp
DeepSeek-V3.2-Exp is an experimental iteration introducing DeepSeek Sparse Attention (DSA) to improve long-context training and inference efficiency while keeping output quality on par with V3.1. It explores fine-grained sparse attention for extended sequence processing.
Benchmark results
| Benchmark | Score | Tags | Source |
|---|---|---|---|
| Aider-Polyglot | 74.5% | self-reported llm-stats | link → |
| AIME 2025 | 89.3% | self-reported llm-stats | link → |
| BrowseComp | 40.1% | self-reported llm-stats | link → |
| BrowseComp-zh | 47.9% | self-reported llm-stats | link → |
| CodeForces | 70.7% | self-reported llm-stats | link → |
| GPQA | 79.9% | self-reported llm-stats | link → |
| HMMT 2025 | 83.6% | self-reported llm-stats | link → |
| Humanity's Last Exam | 19.8% | self-reported llm-stats | link → |
| LiveCodeBench | 74.1% | self-reported llm-stats | link → |
| MMLU-Pro | 85.0% | self-reported llm-stats | link → |
| SimpleQA | 97.1% | self-reported llm-stats | link → |
| SWE-bench Multilingual | 57.9% | self-reported llm-stats | link → |
| SWE-Bench Verified | 67.8% | self-reported llm-stats | link → |
| Terminal-Bench | 37.7% | self-reported llm-stats | link → |