DeepSeek unifies modes into single intelligent model, releases V4.1 flash

DeepSeek has announced the integration of its previous fast, expert, and vision modes into a unified intelligent model, eliminating the need for manual switching. The new model automatically detects query complexity, activates visual capabilities when images are input, and adjusts processing power based on task difficulty. The company also launched the DeepSeek-V4-Flash-Vision-Exp model on its API platform, supporting three call formats. DeepSeek plans to release V4.1 Flash on September 10, 2026, with performance, cost, speed, and total time metrics surpassing V4 Pro. V4 Pro will be phased out on September 14, 2026, with users redirected to V4.1 Flash. A mid-version test of V4.1 Flash showed improved native multimodal support, faster performance, and lower costs. Pricing for Flash models was also updated, with different rates for peak and off-peak hours.