DeepSeek-V4.1 Flash Slashes KV Cache to One-Eighth, Handles 1M Tokens
DeepSeek-v4.1 Flash: Pushing the Limits of KV Cache Compression
DeepSeek-V4.1 Flash, a 552B-parameter multimodal MoE, compresses KV cache to a quarter of its predecessor's runtime footprint and an eighth of its persistent storage. It activates only 8B parameters during prefill and 16B during decode, thanks to a Causal Encoder-Decoder architecture, CSA2 cross-layer compression, and FP4 quantization. The result: near-linear scaling to 1M tokens and roughly 420 tokens per second, making long-horizon agent workflows far cheaper to serve.
When DeepSeek-V4.1 Flash was released, I thought it might just be a post-training iteration version... but after using it for a while, I found it reached nearly 420 Tokens/s in speed, and then Cui said all DeepSeek-V4 Pro models would be taken offline... suddenly I felt this was no small matter...
- arikrahman
I am very impressed with the KV Cache Compression work as well as the prefix cacheing making queries converge on practically free.
- mmastrac
I've been working with an automatic incremental context compactor enabled and it's been surprisingly helpful. It was particularly effective with DS41f - I think I was running at an effective session length of 5M, with the model running around 300k-400k and it was holding on both speed and intelligence.
TBH I also ran the 400tok/s preview and that was just nuts. I just let the thing compact over and over over the course of a day attacking a couple of tough problems
- vivzkestrel
404 on the blog page? https://zartbot.github.io/blog/