RuneBench
Not yet importedCoding-agent tasks inside an emulated RuneScape world, scored by skill progression and resource collection over fixed time windows.
2 task previewsagent · sandboxedReview requiredversion upstream
Complex import. Requires a stateful game sandbox, SDK/MCP control, 20–40 minute leases, numeric rewards, and clean resets.
What it measures
RuneBench asks agents to write and execute TypeScript against a RuneScape SDK and MCP control surface. Numeric rewards and time-series progress are part of the benchmark, not optional telemetry.
A faithful Trunchbull import requires longer sandbox leases, deterministic world resets, readiness checks, SDK/MCP tools, concurrent trials, and durable score and video artifacts.
not yet imported
Not yet imported. This is the recommended first simulation-center integration, but it requires production-quality lifecycle and scoring work.
Tasks
2 previewed| id | task | runtime |
|---|---|---|
| runebench/skill-xp | Skill progressionAccumulate measurable skill experience within a fixed episode.tools: RuneScape SDK, MCP control | sandbox |
| runebench/gold | Gold collectionPlan and execute a resource strategy against a numeric reward.tools: RuneScape SDK, MCP control | sandbox |
Tools
2 requested| tool | description | kind | required |
|---|---|---|---|
| RuneScape SDK | Structured observations and actions for the emulated game world. | sandbox | yes |
| MCP control | Exposes the benchmark's agent-control interface inside the lease. | sandbox | yes |