AIHacker Newsabout 2 hours ago
DeepSeek-V4.1 Flash Achieves 420 Tokens/s
16 min read
DeepSeek released V4.1 Flash, a 552B-parameter multimodal model achieving nearly 420 tokens/s. It uses a Causal Encoder-Decoder architecture to compress KV Cache by 4x, reducing storage to 1/4 of its predecessor for long-context agent workflows.
Level
Hype check
Tap to vote and see what everyone thinks.
In this storyDeepSeek
