Skip to main content

Introducing DeepSeek-V3.2-Exp

πŸš€ Introducing DeepSeek-V3.2-Exp β€” our latest experimental model!

✨ Built on V3.1-Terminus, it debuts DeepSeek Sparse Attention (DSA) for faster, more efficient training & inference on long context.

πŸ‘‰ Now live on App, Web, and API

πŸ’° API prices cut by 50%+!

⚑️ Efficiency Gains​

πŸ€– DSA achieves fine-grained sparse attention with minimal impact on output quality β€” boosting long-context performance & reducing compute cost.

πŸ“Š Benchmarks show V3.2-Exp performs on par with V3.1-Terminus.

πŸ§‘β€πŸ’» API Update​

πŸŽ‰ Lower costs, same access!

πŸ’° DeepSeek API prices drop 50%+, effective immediately.

πŸ”Ή For comparison testing, V3.1-Terminus remains available via a temporary API until Oct 15th, 2025, 15:59 (UTC Time). Details: https://api-docs.deepseek.com/guides/comparison_testing

πŸ”Ή Feedback welcome: https://feedback.deepseek.com/dsa

πŸ›  Open Source Release​

πŸ”— Model: https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Exp

πŸ”— Tech report: https://github.com/deepseek-ai/DeepSeek-V3.2-Exp/blob/main/DeepSeek_V3_2.pdf

πŸ”— Key GPU kernels in TileLang & CUDA (use TileLang for rapid research prototyping!)