In the technical report, why is the performance on Android World different?

[ZhangDailing8]

In the technical report, why is the performance of the various models in Table 7 and Table 8 different? For example, UI-TARS-1.5 scores 64.2 and 14.1 respectively, and other models also show inconsistencies.

[junyangwang0410]

Table 7 shows the performance of the model or framework under different training formats. Table 8 analyzes the performance of VLLM under different agent frameworks.