Compile. Benchmark.
Deploy.

We compare Muna-compiled models against conventional deployments across artifact size, cold starts, time to first token, and decode throughput. Every number below is derived from raw result files you can download, and every run can be reproduced from a public script.