DeepSeek announced that it has merged the original fast mode, expert mode, and image recognition mode into one, so users no longer need to manually switch.
DeepSeek has integrated them into a unified intelligent mode: the model can automatically detect the complexity of the query, activate visual capabilities when detecting image input, and adjust processing power according to task difficulty. Before this, the multimodal visual understanding model DeepSeek-V4-Flash-Vision-Exp was already launched on the DeepSeek API platform, supporting three formats for calling: Chat Completions, Messages, and Responses.
V4.1 Flash Outperforms V4 Pro, Old Models to be Retired on September 14
DeepSeek previously announced that it plans to officially release the V4.1 Flash model around September 10, 2026 Beijing Time. After internal and external testing, V4.1 Flash has comprehensively surpassed V4 Pro in performance, cost, speed, total time, and other metrics.
DeepSeek plans to retire the V4 Pro service at 12:00 Beijing Time on September 14, 2026. At that time, V4 Pro will be routed to V4.1 Flash and charged based on the V4.1 Flash pricing.
According to a notice released by DeepSeek in its official communication group, the intermediate version of V4.1 Flash used a new model structure during internal testing, featuring native multimodal support, stronger capabilities, faster speed, and lower costs.
Flash Series Price Adjustment同步
In addition, DeepSeek also announced that starting at 12:00 on September 10, the Flash series pricing will be adjusted: the unit price for input cache hit during off-peak hours is 0.02 yuan, the unit price for input cache miss is 1 yuan, and the output unit price is 4 yuan; the peak hour price is twice the off-peak hour price.
