DeepSeek 发布 V3.2-Exp：引入稀疏注意力，API 价格下调超 50%

DeepSeek-V3.2-Exp: Boosting Long-Context Efficiency with DeepSeek Sparse Attention DeepSeek-AI research@deepseek.com Abstract We introduce DeepSeek-V3.2-Exp, an experimental sparse-attention model, which equips DeepSeek-V3.1-Terminus with DeepSeek Sparse Attention (DSA) through continued train-ing. With DSA, a fine-grained sparse attention mechanism powered by a lightning in-dexer, DeepSeek-V3.2-Exp achieves significant efficiency improvements in both training and inference, especially in long-context scenarios. The model checkpoints are available at 1. Architecture Compared with DeepSeek-V3.1-Terminus, the last version of DeepSeek-V3.1, the only architec-tural modification of DeepSeek-V3.2-Exp is the introduction of DeepSeek Sparse Attention (DSA) through continued training. [...] The post-training of DeepSeek-V3.2-Exp also employs sparse attention in the same way as the sparse continued pre-training stage. In pursuit of a rigorous assessment of the impact of introducing DSA, for Dee

DeepSeek API Docs Logo DeepSeek API Docs Logo # Introducing DeepSeek-V3.2-Exp 🚀 Introducing DeepSeek-V3.2-Exp — our latest experimental model! ✨ Built on V3.1-Terminus, it debuts DeepSeek Sparse Attention (DSA) for faster, more efficient training & inference on long context. 👉 Now live on App, Web, and API 💰 API prices cut by 50%+! ## ⚡️ Efficiency Gains​ 🤖 DSA achieves fine-grained sparse attention with minimal impact on output quality — boosting long-context performance & reducing compute cost. 📊 Benchmarks show V3.2-Exp performs on par with V3.1-Terminus. ## 🧑‍💻 API Update​ 🎉 Lower costs, same access! 💰 DeepSeek API prices drop 50%+, effective immediately. [...] 💰 DeepSeek API prices drop 50%+, effective immediately. 🔹 For comparison testing, V3.1-Terminus remains available via a temporary API until Oct 15th, 2025, 15:59 (UTC Time). Details: ## 🛠 Open Source Release​ 🔗 Model: 🔗 Tech report: 🔗 Key GPU kernels in TileLang & CUDA (use TileLang for rapi

License ## Introduction We are excited to announce the official release of DeepSeek-V3.2-Exp, an experimental version of our model. As an intermediate step toward our next-generation architecture, V3.2-Exp builds upon V3.1-Terminus by introducing DeepSeek Sparse Attention—a sparse attention mechanism designed to explore and validate optimizations for training and inference efficiency in long-context scenarios. This experimental release represents our ongoing research into more efficient transformer architectures, particularly focusing on improving computational efficiency when processing extended text sequences. [...] DeepSeek Sparse Attention (DSA) achieves fine-grained sparse attention for the first time, delivering substantial improvements in long-context training and inference efficiency while maintaining virtually identical model output quality. To rigorously evaluate the impact of introducing sparse attention, we deliberately aligned the training configurations of DeepSeek-V3

## Introduction We are excited to announce the official release of DeepSeek-V3.2-Exp, an experimental version of our model. As an intermediate step toward our next-generation architecture, V3.2-Exp builds upon V3.1-Terminus by introducing DeepSeek Sparse Attention—a sparse attention mechanism designed to explore and validate optimizations for training and inference efficiency in long-context scenarios. This experimental release represents our ongoing research into more efficient transformer architectures, particularly focusing on improving computational efficiency when processing extended text sequences. [...] DeepSeek Sparse Attention (DSA) achieves fine-grained sparse attention for the first time, delivering substantial improvements in long-context training and inference efficiency while maintaining virtually identical model output quality. To rigorously evaluate the impact of introducing sparse attention, we deliberately aligned the training configurations of DeepSeek-V3.2-Exp wi

-- Listen Share Updated: Novemeber 19th, 2025 DeepSeek’s latest experimental model, V3.2-Exp, is computationally more efficient than its predecessor without sacrificing output quality much. Released on September 29, 2025, this open-weight 685-billion-parameter model introduces a novel DeepSeek Sparse Attention (DSA) mechanism that slashes inference cost and processing time for long inputs, while delivering results on par with its predecessor V3.1-Terminus. Crucially, V3.2-Exp comes with a radical price drop, over 50% cut in API pricing, making high-end AI more affordable. In this review, I’ll cover what makes V3.2-Exp special, how it stacks up against rival LLMs, and what early users and experts are saying about its real-world performance. ## A Leaner Architecture with 128K Context [...] With V3.2-Exp, DeepSeek has delivered an uncommon thing in the AI race: a meaningful architectural innovation paired with immediate real-world benefits. This model is not just a paper experiment;

View organization page for DeepSeek AI 195,075 followers 🚀Introducing DeepSeek-V3.2-Exp — our latest experimental model! ✨Built on V3.1-Terminus, it debuts DeepSeek Sparse Attention(DSA) for faster, more efficient training & inference on long context. 👉Now live on App, Web, and API. 💰API prices cut by 50%+! To view or add a comment, sign in View organization page for DeepSeek AI 195,075 followers 🚀Introducing DeepSeek-V3.2-Exp — our latest experimental model! ✨Built on V3.1-Terminus, it debuts DeepSeek Sparse Attention(DSA) for faster, more efficient training & inference on long context. 👉Now live on App, Web, and API. 💰API prices cut by 50%+! To view or add a comment, sign in View profile for Arnab Dey, PhD [...] ## More Relevant Posts View profile for Ke Zhang DeepSeek just dropped V3.2-Exp, with a fresh sparse attention engine that supercharges long-context conversations—slashing compute while keeping responses crisp. Perfect for multi-turn voice dialogs in robotics,