What is an API base URL, and how is it different from an endpoint?
The base URL is the shared API root your client uses to build requests. The operation endpoint adds a route for a specific action. Your provider's dashboard is an account interface, not necessarily the API host. Copy the root from the provider's integration documentation or a configuration that already works in your application.
The addresses below use the reserved example.com domain and are fictional. They illustrate how to read documentation; they are not working services or a reason to guess your provider's URL.
- Documented base: https://api.example.com/v1 — enter this value when the provider explicitly documents /v1 as the root.
- Full operation: https://api.example.com/v1/chat/completions — if the documented root is /v1, enter the root rather than this full chat operation URL.
- Dashboard: https://example.com/dashboard — an account page is not an API base URL.
- Custom root: https://gateway.example.com/openai/v1 — keep the complete documented prefix; do not replace it with a guessed /v1.
Should the base URL include /v1?
Keep /v1 when it is part of the API root required by the provider and the client you are configuring. Do not add it automatically to every host. Some clients append version or operation paths themselves, so their configuration may differ from ModelVerify's base URL field.
If a request fails, compare the final request path with the provider's documentation. Repeated version prefixes or operation paths can produce a 404. A version prefix describes the API route, not the identity or generation of the language model.
Where can I find the exact model ID?
Use the identifier required in the request's model field. Find it in the provider's API reference, an available-model list, or your application's working configuration. Preserve spelling, case where required, punctuation and version suffixes. A marketing name or model family is not necessarily an accepted request identifier.
For a fictional example, a dashboard could advertise ‘Example Pro’ while the request requires example-pro-2026-10. Copy the documented identifier instead of converting the display name yourself. If a provider exposes a model-list endpoint, it may list only models available to the current key, project or region.
Is the requested model ID the same as the reference model?
No. The requested model ID tells the gateway what to route. The selected official reference tells ModelVerify which supported behavior pattern to compare against. An alias may resolve differently across versions, accounts or time; the label alone does not independently authenticate the backend.
Select the reference you actually intend to check. If no suitable reference is available, do not choose another version and interpret it as an exact identity test. Keep the identity question unresolved and use workload evaluation for capability and reliability questions.
Which protocol and API key should I use?
Choose the documented request format, rather than inferring it from the model's brand. A Claude-branded model offered through an OpenAI-compatible gateway still uses that gateway's OpenAI request format. Use a key issued for the same provider and account, with access to the exact model and enough quota.
Do not put a key in a URL, screenshot or support post. Prefer a restricted key where supported. ModelVerify's web form does not require writing a script, and verification requests may incur provider charges.
What should I check after a 401, 403 or 404?
Review the response together with the provider's documentation: credentials and access for 401/403, and the documented root, operation path and model availability for 404. Providers use status codes differently. Change one field at a time, and never treat a connection failure as a model mismatch.
Once configuration is confirmed, open the verification form and compare the endpoint with the appropriate supported reference. A successful connection establishes reachability; the behavioral report addresses a separate identity question.