AGENT EXPERIMENTS

Agent LabAgent 实验室

We start from real problems and record how different models perform inside the same agent. Pick an experiment to see the results and the evidence behind them.

从真实问题出发，记录不同模型在同一 Agent 下的表现。选一个实验，看结果，也看证据。

Hermes

1 experiment1 项实验

One agent, many models: how each one handles real development tasks.同一个 Agent，观察不同模型如何完成真实开发任务。

EP 01Security fix安全修复

Credential leak fix: after the tests pass凭据泄漏修复：通过测试之后

Starting from a real vulnerability, we compare each model's fix and trace the client-copy gap that remains after the original tests pass.从一个真实漏洞出发，比较模型的修复结果，追查原题通过后仍存在的客户端复制缺口。

9 models
18 main runs
2 extended-budget probes

9 个模型
18 次主实验
2 次延时探针

Read report →阅读报告 →View evidence ↗查看证据 ↗

Coming next即将加入

PIThe next set of agent experiments.下一组 Agent 实验。
More agents更多 AgentMore experiments will be added, grouped by agent.后续实验会按 Agent 持续加入。