Web search

Raon runs web search itself when it is enabled for your organization and configured on the server. Request it per call:

The model decides whether and when to search, and may search several times before it answers. Search and the additional model turns can increase response time. Server limits also apply; a requested max_uses does not override them. Search failures or exhausted limits can lead to an answer without further searches.

The response's usage counts the model's last turn, which already holds the whole conversation and every search result once. Those are the tokens you are billed for, and the input count is the conversation's size for the next request.

completion = client.chat.completions.create(
    model="raon-1-mini-preview",
    web_search_options={"search_context_size": "medium"},
    messages=[{"role": "user", "content": "What changed in the latest PUBG update?"}],
)
message = completion.choices[0].message
print(message.content)
for annotation in message.annotations or []:
    print(annotation.url_citation.title, annotation.url_citation.url)

Chat Completions and ResponsesCopy Anchor Link

• search_context_size sets the results per search: low 3, medium 5, high 10.
• Sources arrive as url_citation annotations. On Chat Completions they are in message.annotations, or in the last chunk's delta.annotations when streaming.
• On Responses each search is a web_search_call output item, and filters.allowed_domains limits results to the sites you list.

MessagesCopy Anchor Link