微软 Decision-1 与 Jev 同场跑购物 agent 决策对比

leopardracer @leopardracer原推发布于 2026.10.10

在 X 上打开原推

Microsoft just embarrassed OpenAI's GPT-6 Sol with a model that can't even write a sentence "Same shopping agent, same 20 decisions in a row, real time with no speedups" Every fork = 1 model call Results: Microsoft Decision-1: ordered in 13.0s for $0.042 per million tokens TypeSafe Jev: ordered in 14.5s at the exact same price, output free OpenAI GPT-6 Sol: stuck on step 8 when the clip ends, needs ~68s and charges $2 in / $10 out per million Satya Nadella's team says Decision-1 is 35x faster than GPT-6 Sol and 200x cheaper on Xbox's 10,000-review test

这条是引用着另一条写的：x.com/leopardracer/status/2108479372675977496