Choose the direct API route
Meta’s overview specifies https://api.meta.ai/v1 as the API base and muse-wonder-1.0 as the regular Wonder model ID. Authentication uses your own provider credential. Begin in the official developer dashboard, enable the billing/access required by your account, and keep the key on your server or in a local environment variable.
Do not paste a key into this site’s calculator, a public repository or browser-delivered JavaScript. A key authenticates billable requests. The calculator needs only workload numbers and performs arithmetic locally.
Read a minimal request before running it
The following uses the official Responses endpoint and regular Wonder ID. It is an instructional example, not a request executed for this guide. Set MODEL_API_KEY in your own shell first; do not replace it in a shared screenshot.
curl https://api.meta.ai/v1/responses \
-H "Authorization: Bearer $MODEL_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"muse-wonder-1.0","input":"Explain a cache hit in one sentence."}'Begin with one short request. Confirm the returned model, output and usage before adding files, tools, parallel workers or a retry loop. This isolates authentication and request-format mistakes from application logic.
Budget the whole request loop
The API pricing page covers the three token rates and separate search charges. Multiply the per-request estimate by every attempted model call, including planner steps, retries and sub-agent calls. A user-visible task can contain several billable requests.
- Log usage after each completed request and associate it with a task ID.
- Distinguish cached input from total input; a repeated prompt is not proof of a cache hit.
- Set application limits on tool rounds, output size and retry count.
- Start with a small test cohort and inspect actual charges before raising concurrency.
Diagnose failures by layer
| Symptom | First check | Avoid |
|---|---|---|
| Authentication failure | Credential and selected provider account | Printing a secret into logs |
| Model not available | Exact catalog ID and account access | Inventing a Full API ID |
| Rate-limit response | Retry-After and team usage | An immediate unbounded retry loop |
| Media ignored | API surface and client attachment support | Assuming the model never supports that input |
| Unexpected bill | All calls, cache usage and tools | Counting only the final answer |
For third-party coding clients or OpenRouter, use the integration guide; provider namespaces and credentials differ. Keep a copy of a redacted request and the error code so that troubleshooting can reproduce the problem without disclosing secrets.
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.