NaceAI 发布 NDI 1.0：文档解析对标 GPT-6 Astra 精度

Dean W. Perkins @deanwperkins原推发布于 2026.10.09

在 X 上打开原推

WHO'S ACTUALLY READING THE DOCUMENTS IN YOUR AGENT PIPELINE? That's the question nobody asked while everyone was building the agent on top. Bad parsing means: → Wrong numbers downstream → Tables silently flattened → Retries that cost more than the doc → An agent confidently wrong NDI 1.0 from @NaceAI is one answer: GPT-6 Astra accuracy on documents, at a tenth of the price.

这条是引用着另一条写的：「NDI 1.0：Nace 发布文档处理小模型」