Skip to content

Benchmarks

Muxlio includes a reference suite for checking workflow execution and saving the raw results behind measurements. It is a tool you can run, not a published claim that Muxlio or one agent provider is faster than another.

What it checks

The suite covers sequential steps, parallel work, large artifacts, cancellation, and recovery. Optional agent cases cover mixed-agent work, coding, and a workflow authored by an agent. Each case has an independent correctness check.

Command-only cases exercise the execution engine without model calls. Agent cases depend on the providers, models, machines, and settings you choose.

What the results mean

Read correctness and completion results along with timing. Missing token or cost data stays unavailable; it is not zero. The suite retains raw observations, plans, events, attempts, and verification results so a measurement can be inspected.

No real-provider performance results are shipped with this suite. To run it, understand its limits, or publish results, use the benchmark methodology.