Benchmark
RuneBench
Test planning and tool use by having coding agents play RuneScape.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
Game-world skill tasks through a TypeScript SDK and emulated server.
Methodology
Agents read game state, consult supplied documentation, and execute actions through code. Skill tasks reward peak experience gain in a 15-second window to encourage strategy discovery.