Interlock: From Local Correctness to Task Completion in Long-Horizon Agents

Open paper

Benchmark	Model	Rank	Results
agents-on-automationbench-public	Qwen3.6-27B (Interlock-RFT)	–	Success Rate: 18.33