MiniCPM5-2B vs. Spark-X2.5-4B / -64% thinking, x1.5 speed while keeping the accuracy of xhigh

We gave two small LLMs a real customer-service task. One got 100/100. The other stopped with the job half-done. I wanted to test something more meaningful than benchmark scores: can a small model actually complete a multi-step agent task without handing the work back to the human? The task: A

中文全文翻译正在补齐，可先查看下方英文原文。

We gave two small LLMs a real customer-service task. One got 100/100. The other stopped with the job half-done. I wanted to test something more meaningful than benchmark scores: can a small model actually complete a multi-step agent task without handing the work back to the human? The task: A