今天，我们 @lobehub 团队发布了一个我觉得很有意义的新Benchmark：Interaction Bench。
This is a vibe coding post classified by AGK Intelligence as Code review & safety (a launch), kept by the Vibe Coding Radar because it carries real work, not commentary.
今天，我们 @lobehub 团队发布了一个我觉得很有意义的新Benchmark：Interaction Bench。 一直以来，我们都觉得Agent的评测和真实用户体验之间，存在着相当大的距离。 真实用户不会告诉Agent哪一行代码出了Bug，也不会给它一份详细的修复方案。用户只会说，我想实现这个功能，或者这个东西坏了，你帮我解决一下。 所以，这次我们尝试换一种方式来评估Agent。 Interaction Bench从25个真实的开源应用中提取了44个任务，只告诉Agent用户想实现的目标，让它自己探索环境、发现问题、尝试解决，并验证最终结果。 首轮结果里GPT-6 Astra以34.1%排名第一，Claude Opus 5.5为 27.3%，其他模型的成功率只有4.5%～15.9%。 即便是目前最强的模型，面对这些真实任务，依然有超过六成无法完成。 我觉得这也是我们做这件事的意义。 随着Agent开始真正进入软件、接管工作流，我们需要更多能够反映真实用户体验的评测。 毕竟对于用户来说，Benchmark上的分数再高，都不如一句「这件事我已经帮你搞定了」。 目前还是0.1 Preview，欢迎大家体验、提 ssue，也欢迎一起来建设。 GitHub：https://github.com/eval-sys/ibench Website：https://ibenc
Posted by Max For AI (44.8k followers) 7 days ago · 96 likes · 12.4k views · view the original post on X. Kept by the Vibe Coding Radar as Code review & safety. Tools mentioned: Interaction Bench.
More vibe coding work like this
- Nous just shipped an official Security plugin for Hermes… it reviews a local repo on the… — @HermesAgentTips
- OpenAI models won't let me protect myself. — @marclou
- Do you know what your coding agent has run? — @hqmank
- Your playtesters have competition. — @rryssf
- first time i've used a cat emoji on the timeline — @benhylak
- Devin Review 2.0 cuts the cost of a PR review by 82% while finding more bugs per PR and… — @devindevelopers
- Introducing Devin Review 2.0: the best way to review PRs with Devin. — @cognition
- Protip working with AI coding agents. — @PovilasKorop
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 7.9k posts from 5k X accounts over the last 14 days, 2.6k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-06 15:18 UTC. Full method.