Research · AI-agent task-completion benchmark and results dataset
METR Time Horizons: coverage and access
agent capability progression, long-task automation analysis and model evaluation
What it covers
Geography and granularity
Geography: frontier AI models evaluated on software/research tasks
Granularity: model, task, human task duration, success rate and fitted time horizon
Source type: AI-agent task-completion benchmark and results dataset
Potential use
agent capability progression, long-task automation analysis and model evaluation
Access and cost
interactive chart plus YAML benchmark/result downloads
none for public data
Cost: free
Update cadence: new evaluation releases; downloadable versioned results
- Provider page ↗
- Documentation and access details ↗Check the provider's current terms and instructions before use
Interpret carefully
Limits that affect the decision
suite coverage is limited, long horizons are unreliable and time horizon measures task difficulty rather than autonomous operating time
new evaluation releases; downloadable versioned results
This record summarizes source documentation; Jon Bishop did not conduct an independent audit of the underlying dataset.