Benchmarks

Benchmark takes an environment and a set of models. Report is the scored result — use it to compare candidates before you pin a route.

from enroute import Benchmark, Enroute, Environment, Report

env = Environment(name="math-smoke", version="0.1.0")
client = Enroute()
benchmark = Benchmark(env, ["openai/gpt-5.6-luna"], client=client)
report: Report = benchmark.run()
print(report.environment, report.win_rates)
print(report.to_markdown())

The Benchmarks tab in this product will host those reports for the workspace. The page is a placeholder until the runner is wired to hosted keys.