Venice AI 去审查版 V4.1 Flash 实测：代码 454 tok/s

Neko Legends @softpoo原推发布于 2026.10.24

在 X 上打开原推

Average 454 tok/s Deepseek V4.1 1017 single stream code, 175 tok/s prose. The model from Venice AI with model "abliteration-abliterated-model-large-flash" is actually the fast version of Deepseek V4.1 1017 that is abliterated. It costs twice as much but also twice as fast, which is still less than half the cost of GLM 5.3