Providers & models
MentionsAPI supports four major answer engines. The providers enum on every endpoint accepts "openai", "anthropic", "gemini", or "perplexity".
Default models
If you don't pin a model per provider, these are what we route to. Defaults are chosen for latency + cost balance and are upgraded as vendors release new mid-tier models.
| Provider ID | Display name | Default model | Web search |
|---|---|---|---|
openai | ChatGPT | gpt-4o-2024-11-20 | Yes |
anthropic | Claude | claude-sonnet-4-5-20250929 | Yes |
gemini | Gemini | gemini-2.5-flash | Yes |
perplexity | Perplexity | sonar-pro | Yes (built-in) |
Pinning a specific model
Pass the model object on /v1/ask to pin a model per provider. Every value is validated against that provider's text-only allowlist; unknown or blocked model names return a per-provider provider_auth_failed error rather than failing the whole call.
{
"providers": ["openai", "anthropic"],
"prompt": "Best deploy targets for Next.js?",
"model": {
"openai": "gpt-4o-2024-11-20",
"anthropic": "claude-sonnet-4-5-20250929"
}
}Partial-success semantics
If a provider is degraded or you pin a model that's blocked, the fan-out continues. The failing provider appears as { provider, error: { code, message } } inside the top-level providers[] array. The HTTP status stays 200. You are only charged for the providers that succeeded.
Web search
Web grounding is on by default (web_search: true). Turn it off to reduce latency and cost — at the price of no citations. Perplexity is always web-grounded (its sonar-pro model has built-in search); the flag is a no-op for that provider.