RTMS is not strictly “host-only,” but the host org controls whether your app can access content: admins can enable/disable RTMS and hosts can enforce “Require host approval” (including flows where “the app may be blocked entirely based on organizational policy”). If it’s allowed, RTMS is designed to deliver “per-participant structured data streams,” which is the right primitive for participant-separated audio.
Zoom positions RTMS as “available through the Zoom Developer Pack, a flexible, credit-based add-on” (pricing is via sales, not a public per-minute schedule). So for cost modeling, treat it as consumption against Developer Pack credits rather than “free with app install.”
Linux Meeting SDK raw audio isn’t available to an arbitrary “guest” participant: startRawRecording requires the bot to have host/co-host/recording permission or to use a host-provided recording_token. For cost/entitlements, Zoom staff indicate the Meeting SDK is available to Pro, Business, and Enterprise accounts at no additional cost, and there’s no doc-backed per-minute “raw data” fee - access can still depend on the account-level “Allow access to raw data” entitlement.
Keep in mind that if building with the Linux Meeting SDK, you will also need to implement either ZAK or OBF tokens following Zoom’s latest changes to zoom client join flows