The DeepSeek Harness v0.1.0-rc.8 version is now available, bringing 14 adjustments in this update, with a focus on enhancing multimodal capabilities by supporting native image requests and mixed text-image inputs. Commands such as /goal and /plan can now receive images, while the @menu has added file and conversation reference functions. The sub-agent system has been expanded, allowing users to install Claude Code and Codex as needed. The Windows terminal experience has been optimized, addressing issues with image requests, streaming generation, and custom gateways. Additionally, the new version has built a tool layer for text-only models to process visuals, converting images into structured information by utilizing tools such as OCR, color statistics, and pixel scanning for text model inference. Meanwhile, the update has also strengthened the plugin approach, incorporating optimizations like web_search concurrent queries to enhance overall performance.