The highlight of China’s AI industry, DeepSeek-V4, has just open-sourced, directly generating over 1 million words of ultra-long context!


DeepSeek-V4 introduces a brand-new attention mechanism and compresses at the token level, combined with DSA sparse attention (DeepSeek Sparse Attention), reducing computational power and VRAM consumption when processing extremely long contexts.
Test data shows that in the Agentic Coding evaluation, V4-Pro's performance has surpassed the industry-standard Sonnet 4.5 and is close to the level of Opus 4.6 in non-thinking mode.
To meet different market demands, DeepSeek has implemented a dual-line strategy: the Pro version focuses on complex reasoning and world knowledge, while the smaller parameter Flash version is responsible for providing ultra-low-cost API calls.
View Original
post-image
post-image
post-image
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pin