From 23 August Beijing time, DeepSeek will charge its off-peak API rate all weekend, so Saturday batch jobs fall on the cheaper half-price band. It also released an experimental Flash model that reads images through the API.
DeepSeek said that from the start of 23 August Beijing time it will charge its off-peak API rate for all of Saturday and Sunday, dropping the peak split it introduced a week earlier for those two days. Off-peak is half the peak rate. As IT之家 reported, the weekend rule extends the cheaper band from a few hours to two full days, so batch jobs and evaluation runs scheduled at the weekend now fall entirely inside it.[1]
DeepSeek split its API pricing into peak and off-peak bands on 16 August, set off-peak at half the peak rate and said the aim was to spread load through the day. The V4-Pro output price reported by IT之家 falls from 27 yuan to 13.5 yuan per million tokens.[1]
DeepSeek published DeepSeek-V4-Flash-Vision-Exp on its developer platform, an experimental model that adds image understanding to V4-Flash. The company says text performance is unchanged and that multimodal agent scores come close to Opus-4.8; that comparison is the vendor's own, with no independent test behind it. Access is through the API only, with model='deepseek-v4-flash-vision-exp'; the weights stay unpublished. Images are billed as tokens, at most 384 per image, at V4-Flash rates, and the release adds a Files API for uploads. Coverage put the ceiling at 600 images per request and noted compatibility with OpenAI's Chat Completions and Responses interfaces and Anthropic's Messages endpoint.[2]
