Jev 1.13 实测：只做判断不写文本的决策模型

Zocker @ZockerAlt原推发布于 2026.10.07

在 X 上打开原推

Saw the NEAR AI Cloud note and had to check what Jev 1.13 actually does before writing it off as another model drop. TypeSafe built it to judge, not to write, so you send context and a few typed questions and get probabilities back. No paragraphs to clean up, no JSON to coax out of a chat model. That shows up in places like: • sending a ticket to the right queue • tagging messy input without a custom parser • deciding whether a score is high enough to act on They priced input at $0.042 per million tokens and left output free. Worth a test if your stack spends more time formatting answers than making them.