DeepSeek-v4.1 Flash: Pushing the Limits of KV Cache Compression
DeepSeek-V4.1 Flash is a 552B-parameter multimodal MoE model with 1M-token context achieving 4x KV cache compression for long-horizon agent workloads.
A detailed analysis of the DeepSeek-V4.1 Flash technical report describes a 552B-parameter multimodal mixture-of-experts model supporting contexts up to 1 million tokens. Its Causal Encoder-Decoder (CED) architecture activates 8B parameters during prefill and 16B during decode, and reportedly delivers about 420 tokens/s. Joint optimization of architecture (CSA2 cross-layer compression), FP4 KV cache precision, and deployment strategy cuts runtime KV cache to roughly 1/4 and persistent KV cache to about 1/8 of DeepSeek-V4-Flash at the same sequence length, targeting storage and bandwidth bottlenecks in long-horizon agent serving. The author notes all DeepSeek-V4 Pro models were taken offline following the release.
- 552B-parameter multimodal MoE with up to 1M-token context
- CED architecture activates 8B (prefill) / 16B (decode) parameters per token
- KV cache reduced to ~1/4 runtime and ~1/8 persistent vs DeepSeek-V4-Flash
- Uses CSA2 cross-layer compression, FP4 KV cache, and sparse attention optimizations
- Reported ~420 tokens/s; DeepSeek-V4 Pro models taken offline
- Tags
- deepseekdeepseek-v4.1-flashfp4inference-optimizationkv-cachelong-contextmixture-of-expertssparse-attention
- Vendors
- DeepSeek
- Organizations
- DeepSeek
- AI models
- DeepSeek-V4.1-FlashDeepSeek-V4-Flash
From the original article
13 min read at zartbot.github.ioWhen 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...
By mfiguiere · zartbot.github.io
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.