VideoGameBench
Not yet importedVisual computer-use evaluation across Game Boy, DOS, and browser games, with a paused Lite mode for latency-fair comparison.
1 task previewagent · sandboxedMixed assets / review requiredversion lite
Complex import. Requires visual computer use, game-state reset, recordings, and asset-aware sandbox provisioning.
What it measures
VideoGameBench measures visual perception and control against recognizable games. Lite mode pauses the environment while the model reasons, reducing the degree to which provider latency determines the score.
A public Trunchbull import must separate paused and real-time results and must never redistribute proprietary game assets. Commercial assets need an operator- or user-supplied path.
not yet imported
Not yet imported. Start with Lite mode and only environments whose assets can be legally provisioned.
Tasks
1 previewed| id | task | runtime |
|---|---|---|
| videogamebench/lite-browser | Paused browser gameObserve a frame and act while the environment waits for model reasoning.tools: screen capture, keyboard control | sandbox |
Tools
2 requested| tool | description | kind | required |
|---|---|---|---|
| screen capture | Provides the current visual observation. | sandbox | yes |
| keyboard control | Sends bounded game inputs to the active environment. | sandbox | yes |