LiteLLM Model Catalog API
Look up model prices, context limits, and capabilities. Public access, no API key.
curl --fail-with-body 'https://api.models.litellm.ai/v1/models/gpt-4o'
Choose a lookup
| Lookup | Request |
|---|---|
| Exact LiteLLM model string | GET /v1/models/{model_id} |
| Models from a provider | GET /v1/providers/{provider}/models |
| Exact provider-relative model name | GET /v1/names/{name}/models |
| Available providers | GET /v1/providers |
| Snapshot and generation time | GET /v1/snapshot |
Names and IDs are case-sensitive. Keep slashes in model strings; URL-encode other reserved characters. A name removes only the matching provider prefix, not version suffixes. Query parameters do not filter these static responses.
curl --fail-with-body 'https://api.models.litellm.ai/v1/providers/anthropic/models' curl --fail-with-body 'https://api.models.litellm.ai/v1/names/gpt-4o/models'
Responses and pagination
Responses have data and meta. A model lookup returns one object. Provider/name lookups return up to 50 objects; follow links.next until it is null. Next links identify the same snapshot. If an older snapshot returns 404 after a deployment, restart from the original lookup.
Each model exposes id, name, provider, mode, context, pricing, and capabilities. The token-price summary uses USD per million tokens. Null means unknown; zero means free. pricing.source_rates preserves the source's cost fields, including non-token, cache, batch, and tiered rates. The summary is not a complete billing formula.
Freshness
Data comes from LiteLLM's public cost map and is rebuilt on deployment. meta.generated_at is the build time, not the date the provider changed its prices. Compare meta.snapshot to identify changes. There is no bulk price-map download endpoint.
Limits and errors
The API firewall limits requests to 120 per minute per IP. It is a regional fixed-window abuse limit, not an account quota; users behind shared networks share the allowance. If you receive HTTP 429, wait at least a minute before retrying. Automatic DDoS protection may separately reject or challenge traffic. Unknown resources or expired snapshots return 404. Supported methods are GET, HEAD, and OPTIONS. Browser clients can use CORS.
Need a higher limit? Choose Give feedback on the model catalog website. Include “API limit request,” your expected requests per minute, your use case, and a reply address. Higher limits require review.
We measure API request counts and cache status. Usage analytics exclude IP addresses, emails, client user agents, query strings, and raw model paths. These counts do not identify unique people.