Gemini 4 Argon 企业文档任务实测：准确率 85% 对 3.1 Pro 71%

Box @Box原推发布于 2026.10.14

在 X 上打开原推

Gemini 4 Argon from @GoogleDeepMind is built for complex, multi-step work. We ran it against Gemini 3.1 Pro on real enterprise document tasks: pull the source files, reason across them, produce the actual deliverable. → Overall accuracy: 85% vs 71% (+14 points) → Data analysis: 93% vs 70% (+23 points) → Life sciences: 88% vs 61% (+27 points) → Healthcare: 87% vs 57% (+30 points) The gains land where the work has to be carried all the way through: reading evidence out of figures, covering the whole batch, and acting on its own findings.

这条是引用着另一条写的：「Google DeepMind 宣布 Gemini 4 Argon 全球上线」