jevPOST
Lightfield 实测 Jev 做客服对话文本理解：准确率追平 Opus 5，成本低 40–150 倍
Lightfield 首席 AI 官在关闭扩展思考的情况下基准测试 Jev，客服对话等文本理解任务准确率与 Opus 5 持平，成本低 40 到 150 倍，并说明将如何接入自家平台。
Lightfield
@lightfld2026.09.22收录原推发布于 2026.09.22
Lightfield @lightfld原推发布于 2026.09.22
Lightfield's Chief AI Officer @vesko_st benchmarked TypeSafe's Jev model on text understanding, including customer-conversation tasks. With extended thinking turned off, Jev matched Opus 5 on accuracy at 40 to 150 times lower cost. Full write-up, including initial thinking on how we'll be using Jev in our own platform: https://lightfield.app/blog/testing-typesafe-jev-on-text-understanding
x.com/lightfld/status/2102149497363354096采集于 2026.09.22 10:12