88Z AI, 24 April — DeepSeek has released and open-sourced the preview of DeepSeek-V4, featuring Pro and Flash variants that both support a 1M context window. The Pro version matches top closed-source models in performance, while the Flash version offers similar reasoning capabilities at a lower cost. The new architecture significantly reduces computation and memory overhead for long contexts.
At a glance
- DeepSeek-V4 preview is open-sourced with Pro and Flash versions supporting 1M context
- New architecture uses DSA sparse attention to significantly reduce long-text computation and memory costs
The update
DeepSeek has officially launched the preview version of its new DeepSeek-V4 model series, releasing it as open source simultaneously. The release includes two specific variants, deepseek-v4-pro and deepseek-v4-flash, both of which support a 1M ultra-long context window. The V4-Pro matches top-tier closed-source models in agent coding, world knowledge, and reasoning performance, while the V4-Flash aims to provide comparable reasoning capabilities at a lower cost. Technically, the series adopts a new attention mechanism combined with DSA sparse attention, significantly reducing computational complexity and memory usage for long-context processing.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- DeepSeekSource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.