[AINews] DeepSeek V4 Pro (1.6T-A49B) and Flash (284B-A13B), Base and Instruct — runnable on Huawei Ascend chips

DeepSeek released V4 Pro (1.6T-A49B) and Flash (284B-A13B) models in both base and instruct variants, with optimizations enabling them to run on Huawei Ascend chips. These releases represent updates to DeepSeek's model lineup with improved parameter efficiency and hardware compatibility tailored to Chinese chip infrastructure. The announcement was covered by Latent Space, a publication focused on AI developments and industry news.

More of my notes on DeepSeek V4 - the really big news is the pricing: both DeepSeek-V4-Flash and DeepSeek-V4-Pro are the cheapest models in …
DeepSeek V4 Pro costs 1.74/1M input tokens and 3.48/1M output tokens, while V4 Flash costs 0.14/1M and 0.28/1M; both models are the cheapest in their class (Simon Willison/Simon Willison's Weblog)
Huawei says its Ascend supernode based on the Ascend 950 AI chips will fully support DeepSeek V4, as DeepSeek launches a preview of its V4 model (Reuters)
DeepSeek releases its new flagship models V4 Pro and V4 Flash in preview, saying V4 Pro trails the performance of state-of-the-art models by about 3 to 6 months (Bloomberg)
DeepSeek V4 Pro has 1.6T total parameters, its largest model by the metric, and V4 Flash has 284B parameters; both models have a context window of 1M tokens (Vincent Chow/South China Morning Post)
huggingface: https://huggingface.co/collections/deepseek-ai/deepseek-v4.

Source: Latent Space | 2026-04-25

Related