DeepSeek-V3.2-Exp is an experimental large language model released by DeepSeek as an intermediate step between V3.1 and future architectures. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism...
Tested on 14 benchmarks with 53.2% average. Top scores: Chatbot Arena Elo — Overall (1422.8%), Chatbot Arena Elo — Coding (1285.5%), Fiction.LiveBench (83.3%).
DeepSeek V3.2 scores 58.7 (100% as good) at $0.25/1M input · 7% cheaper
Multi-language code editing from Aider. Tests editing ability across Python, JavaScript, TypeScript, Java, C++, Go, Rust, and more.
Regularly refreshed coding problems that avoid data contamination. New problems added monthly to prevent memorization.
Unusual and adversarial machine learning challenges. Tests robustness of reasoning about edge cases in ML systems.
Regularly refreshed reasoning problems testing logical deduction, spatial reasoning, and analytical thinking.
Fresh data analysis tasks testing ability to interpret tables, charts, and statistical data.
Regularly updated math problems that test numerical reasoning, algebra, calculus, and combinatorics.
- Typetext
- Context164K tokens (~82 books)
- ReleasedSep 2025
- LicenseOpen Source
- StatusActive
- Cost / Message~$0.001