DeepSeek V3.2 and V3.2-Speciale are now available in Cline

DeepSeek-V3.2 and its variant DeepSeek-V3.2-Speciale, now available through Cline, are designed to enhance agentic workflows by integrating reasoning into tool execution, rather than treating it as a separate step. The V3.2 model features DeepSeek Sparse Attention (DSA) for efficient attention in long-context scenarios and maintains reasoning continuity across multiple tool executions, making it ideal for iterative tasks like coding. Trained on a vast array of synthesized environments and agent instructions, the models are capable of generalizing across different tool-use scenarios, with V3.2 providing a balanced approach for everyday tasks and V3.2-Speciale offering deeper reasoning for complex problems. Both models have been extensively post-trained, including significant reinforcement learning, to perform at a high level on reasoning benchmarks. While they share pricing and a 131K token context window, they do not support images, browser use, or prompt caching. Users can choose between the two models for different needs via the Cline provider dropdown and are encouraged to share their experiences on community platforms.