Open WebUI builds the chat model picker from the model providers connected to the server. When a provider adds a model, refreshing the model list makes the updated catalog available to accounts with access.
The refresh request clears the cached provider model lists before fetching them again. It does not download models or grant access. Newly discovered models without configured access are visible to administrators; standard users still need permission to use them.
An existing OpenAI-compatible connection, an enabled Open WebUI API key, curl, and jq are required. An administrator account can check discovery before the new model has been shared. The model must already be advertised by the connected provider; refreshing cannot repair an unavailable provider or an incorrect connection URL.
Steps to refresh the Open WebUI model list:
- Enter your existing Open WebUI API key at the hidden terminal prompt.
$ read -r -s OPEN_WEBUI_TOKEN
The input stays hidden until Enter is pressed. The key belongs to Open WebUI, not the provider. The header is passed through standard input below so the key is absent from the command arguments and shell history. API keys must be enabled for the account; endpoint restrictions, when enabled, must allow /api/models.
Related: How to enable API keys in Open WebUI - Inspect the model IDs currently returned by Open WebUI.
$ printf 'Authorization: Bearer %s\n' "$OPEN_WEBUI_TOKEN" | \ curl --fail-with-body --silent --show-error \ "https://openwebui.example.com/api/models" --header @- | jq -r '.data[].id' company-chat company-task company-chat-alt
The missing model in this list is company-chat-new. A normal model-list request can return cached data.
- Confirm that the connected provider advertises the missing model ID.
$ curl --fail-with-body --silent --show-error \ "https://llm.example.com/v1/models" | jq -r '.data[].id' company-chat company-task company-chat-alt company-chat-new
The URL must match the existing connection under user menu → Settings → Admin → AI → Connections. If the provider requires authentication, its own credential is required for this request. A fixed Model IDs list in the connection limits discovery even after a refresh.
Related: How to connect an OpenAI-compatible provider to Open WebUI - Refresh the Open WebUI model list with the authenticated API request.
$ printf 'Authorization: Bearer %s\n' "$OPEN_WEBUI_TOKEN" | \ curl --fail-with-body --silent --show-error \ "https://openwebui.example.com/api/models?refresh=true" --header @- | jq -r '.data[].id' company-chat company-task company-chat-alt company-chat-new
refresh=true invalidates the model-list caches for a fresh fetch. The returned IDs still follow the calling account's access permissions. A missing model for a standard user can require an administrator's access grant rather than another refresh.
- Reload the chat tab for the same Open WebUI account.

- Open the model picker beside the message input.

- Select the newly discovered model from the picker.
The selected model name now appears beside the message input. Discovery and selection confirm the refreshed list; they do not test whether the provider can generate a response.