Separate the three roles
The developer publishes or maintains the model. The inference provider serves requests on infrastructure. The API channel is the service through which you access an offer. A channel may route a request among multiple inference hosts, so a channel-level price is not necessarily a pinned-host quote.
Read exchange tags
The exchange uses @developer for a developer, $model for a model and :channel.provider to identify an offer when a host is known. These tags distinguish a model’s identity from its route to inference. Automatic routing remains explicitly labeled rather than being presented as a guaranteed host.
Compare the service as well as the model
Check supported parameters, region, retention, context, quantization, tier and source-reported performance. Two offers for a similarly named model can differ operationally. Medusa links reuse existing model integrations; they do not by themselves guarantee a particular inference host.