As the one-month mark approaches since the official release of V4 Pro, DeepSeek is preparing to begin internal testing for the new V4.1 version. This upgrade brings many highlights, including the much-anticipated price reduction.
DeepSeek's official assistant has announced that the intermediate version of V4.1 Flash is now open for beta testing applications.
Regarding the new version, DeepSeek stated that V4.1 Flash adopts a new model architecture, offering native multimodal support, enhanced capabilities, faster speed, and lower costs.
To call the API, simply keep the base_url unchanged and set the model name to deepseek-v4.1-flash-expires-on-0910. The current pricing is the same as deepseek-v4-flash, with a limit of 20 concurrent requests per account.
This strategy mirrors the previous V4 official release, launching the Flash model first. The release of V4 Flash-0731 was impressive at the time, with performance directly catching up to the level of Opus 4.8.
However, the official V4 Pro release on August 13 was underwhelming. Its performance didn't meet expectations, showing little improvement over V4 Flash-0731, and the subsequent significant price hike further damaged DeepSeek's reputation, turning "Liang Sheng" (Saint Liang) into "Liang Zi" (a derogatory nickname).
Now that V4.1 Flash is entering testing, if it delivers on the promised improvements in strength, speed, and cost—especially with a significant price cut and native multimodal support—it could be highly valuable if its performance rivals models like GPT-5.6 Sol or Opus 5. Let's hope "Liang Zi" becomes "Liang Sheng" again.