Do not mix provider namespaces
| Route | Model identifier observed | Account and bill |
|---|---|---|
| Direct Meta Model API | muse-wonder-1.0 | Meta Model API credential and billing |
| OpenRouter catalog | meta/muse-wonder | OpenRouter credential and billing |
| OpenCode using Meta | Provider configuration chooses the direct ID | Whatever provider you configured |
OpenRouter’s own current catalog lists Muse Wonder and the provider-specific slug. That verifies a catalog entry, not a promise of continuous availability in every routing region. Read the current provider and endpoint information before deploying a workload.
Connect a coding client deliberately
Meta’s coding-agent documentation covers OpenCode and distinguishes its Responses adapter from the Chat Completions adapter. The former supports reasoning replay; the latter has different continuity and attachment behavior. Follow the current configuration for the surface your installed client uses.
Before changing a working client, save its configuration, name the new provider clearly and keep its key in the credential mechanism or environment variable expected by the client. Start a fresh session with a small task and confirm the model actually selected. Do not infer provider choice from a display name alone.
Keep subscriptions and API invoices separate
A Muse Code subscription does not establish an allowance for an arbitrary OpenCode provider configuration. Meta documents additional API keys as pay-as-you-go. OpenRouter also has its own account, routing and billing rules. Keep each route’s invoice and request logs identifiable.
Published model tariffs help create an initial estimate, but routing, cache reporting, tools and repeated calls affect the actual workload. Compare the same accepted task and record every model invocation, rather than comparing two screenshots with different reasoning settings.
Integration troubleshooting checklist
- Identify the client version, provider, base URL and exact model ID.
- Confirm that the credential belongs to the selected provider.
- Match the API surface and adapter to the features you need.
- Inspect the returned model and usage on one small request.
- Check how tool calls, media attachments and conversation state are represented.
- Only then add parallelism, long context and automated retries.
This guide did not execute paid calls through either route or benchmark their reliability. Use the direct API guide to isolate a request-format problem, or Muse Code if you want the first-party agent without a third-party provider configuration.
Sources & evidence
Editorial check: 17 October 2026. Product facts come from the following primary sources; planning checklists and examples are our explanations. No paid inference, model benchmark or regional checkout was run for this guide.