On October 17, 2026, DeepSeek unveiled its V4.1 Flash-1017 large-scale model, accompanied by concurrent updates to its official web portal, APP, and API models. The latest iteration showcases notable enhancements in performance during benchmark evaluations, while simultaneously streamlining task execution times and output lengths. This advancement represents a Pareto improvement, optimizing both performance and cost-efficiency. The new model incorporates 'wish-based' programming capabilities, facilitating the creation of comprehensive projects from the ground up in domains such as game development and 3D modeling, leveraging self-feedback reinforcement learning technology. Furthermore, DeepSeek has made its proprietary search API publicly available, with both Search and Fetch interfaces priced at a mere 5 yuan per 1,000 queries, enhancing accessibility and affordability for users.