PANews reported on April 24th that, according to the Science and Technology Innovation Board Daily, the preview version of DeepSeek-V4 was launched and open-sourced today, including DeepSeek-V4-Pro and DeepSeek-V4-Flash. According to the official DeepSeek documentation, DeepSeek V4 has simultaneously validated its fine-grained expert parallel (EP) solution on NVIDIA GPUs and Huawei Ascend NPUs. Furthermore, the official API page mentions that due to limitations in high-end computing power, the service throughput of the V4-Pro model is currently limited. It is expected that the price of Pro will drop significantly after the Ascend 950 supernode is mass-produced and available in the second half of the year.