ServiceNow Open-Sources AgentHorizon Benchmark for LLM Judges

ServiceNow open-sourced AgentHorizon, a benchmark with 1373 long-horizon tasks and 166 hours of human GUI trajectories, testing whether LLM judges can accurately assess if 100-300 step computer-use tasks are truly completed.

2026-10-10 ~ 2026-10-10 · 2 related posts

• AgentHorizon: A Benchmark Testing Whether LLM Judges Can Verify 100-300-Step Computer-Use Tasks — xhluca · 2026-10-10
• ServiceNow Open-Sources AgentHorizon: 1,373 GUI Trajectories From 166 Hours to Test Agent Judges — xhluca · 2026-10-10