Can not use model's default chat template with tools, the evaluation result maybe can not aligned.

[LiuLinyun]

[chenchen0103]

Thank you for your feedback! In our current benchmark design, the agent and special data types are indeed driven by specific prompts to guide the model's reasoning. On the other hand, normal data types are typically evaluated through text generation, with the capability to integrate remote services or tools (such as API calls).

Regarding your suggestion for native Function Calling (FC), we plan to implement this feature in an upcoming update. This will enhance the support for normal data types, enabling them to interact correctly with external tools.