Control
Models
Bring candidate run evidence and compare accepted outcomes, cost, and task constraints using the same evaluation criteria.
Last updated August 22, 2026
Compare models on work you would accept.
Use cases
Use the same standard
Judge candidate results against a common task and acceptance contract.
Count the complete outcome
Review reported cost alongside acceptance, rather than treating a cheap failed run as a win.
See comparison limits
Identify runs that are missing evidence or cannot fairly be compared.
How it works
- 1
Set the acceptance criteria
Define the task, mandatory requirements, and constraints.
- 2
Bring candidate results
Supply the run evidence and reported costs from your model trials.
- 3
Review the comparison
Inspect accepted outcomes and excluded candidates before selecting a model.
Quickstart
Open Models in the console, provide the required inputs, and start an organization scoped run.
Inputs
- task contract
- candidate models
- evaluation rubric
- budget
Outputs
- ranked candidates
- failure traces
- cost forecast
- routing decision
MCP
Call Models from a compatible agent with the preferred public tool name.
The previous internal service identifier remains accepted for compatibility, but new integrations should use this product name.