Hi Zhipu team, thank you so much for open-sourcing such impressive models and sharing your research!
I had 2 quick questions:

1. How can the BrowseComp and HLE evaluation results be replicated? I noticed the two files "trajectory_search.json" and "glm_4.6_tir_guide.md" but are these enough to reproduce the results?
2. Is the search-agent framework you used for BrowseComp/HLE evaluation open-source, or do you plan to open-source it?

Also, if I can also ask, what the agent framework was used to achieve the SWE-Bench Verified results?

Thanks again for your great work! 🙏

嗨，智谱团队，非常感谢你们开源如此优秀的模型并分享研究成果！
我有两个简短的问题：

1. BrowseComp 和 HLE 的评测结果该如何复现？我注意到仓库里有两个文件 “trajectory_search.json” 和 “glm_4.6_tir_guide.md”，但它们是否足以复现你们报告的结果呢？
2. 用于 BrowseComp/HLE 评测的 搜索代理框架 是否已经开源，或者未来有开源计划吗？

另外，如果可以的话，我还想请教一下，你们在 SWE-Bench Verified 上取得结果时使用的代理框架是什么？

再次感谢你们的卓越工作！🙏