DeepSeek V4.1 Flash 1017 hits 675.9 tok/s locally

Md Ismail Šojal 🕷️ @0x0SojalSecOriginally posted 2026.10.22

Open original on X

Crazy, Deepseek-v4.1-flash-1017 just hit 675.9 tokens per second local Run. the speeds most people still think are impossible. Day by day, inference optimizations are reaching like crazy levels.

This post quotes another post: “DeepSeek V4.1 Flash 1017 measured at 675.9 tok/s at concurrency 1”