vLLM v0.30.0 Released with DeepSeek-V4.1-Flash and MXFP8 KV Storage

Core summary

One sentence to understand this update

vLLM v0.30.0 has been released, featuring support for new models like DeepSeek-V4.1-Flash and storing the entire KV in MXFP8 via FlashMLA V4.1.

Impact & opportunity

What this could mean

This update allows builders to use cutting-edge models more efficiently with vLLM, potentially reducing inference costs and improving performance.