DeepSeek 推出 V4.1 Flash 高吞吐推理服务

Jai @jai_prasad17原推发布于 2026.10.21

在 X 上打开原推

Two cost collapses before lunch. DeepSeek put out a high-throughput inference service for V4.1 Flash, and the detailed writeup only exists on Zhihu Jim Fan open-sourced Flora, a robotics data engine he says matches Astra at 125x lower cost Every time a model call gets cheaper, the list of things worth running an agent on gets longer.