deepseek-v4.1POST
157 题实测：V4.1 Flash 快照关思考得 128 分，数学降到 30 题对 12
作者用 Hermes agent 在关闭思考模式下跑完 157 道测试，得 128 分比九月高 1 分，但数学只对 12/30；并指出只有 effort none 才真正关闭思考。
Mike Gannotti 🔳
@MichaelGannotti10 天前收录原推发布于 2026.10.18
Mike Gannotti 🔳 @MichaelGannotti原推发布于 2026.10.18
Nemo, one of my Hermes agents, ran the new DeepSeek V4.1 Flash snapshot through all 157 tests with thinking off. It scored 128, one up on September, but math fell to 12 of 30. And only effort none actually turns thinking off. https://x.com/MichaelGannotti/status/2111504107844559243
这条是引用着另一条写的：x.com/michaelgannotti/status/2111504107844559243
x.com/michaelgannotti/status/2111504236047654987采集于 2026.10.18 16:30