Web Search

Not yet available. The web search server tool is on the roadmap. The page below documents the intended behavior.

The web search server tool gives the model the ability to issue search queries and read results during a response. No client-side code is needed — OpenDunes handles the retrieval on the model's behalf.

Enabling Web Search#

Add the built-in web_search server tool to your tools array:

{
  "model": "openai/gpt-5.4",
  "input": "What is the current USD to DZD exchange rate?",
  "tools": [
    { "type": "web_search" }
  ]
}

The model decides when to search. It issues one or more queries, reads the results, then synthesises an answer with citations.

Response Format#

When the model searches, the output array includes a web_search_call item followed by the assistant's message:

{
  "output": [
    {
      "type": "web_search_call",
      "id": "ws_abc",
      "queries": ["USD DZD exchange rate today"]
    },
    {
      "type": "message",
      "role": "assistant",
      "content": [
        {
          "type": "output_text",
          "text": "As of today, 1 USD ≈ 134 DA...",
          "annotations": [
            {
              "type": "url_citation",
              "url": "https://...",
              "title": "..."
            }
          ]
        }
      ]
    }
  ]
}

Combining with Other Tools#

Web search can coexist with function definitions in the same tools array. The model chooses which tool to invoke based on the query.

Pricing#

Web search is billed as additional input tokens — each search result snippet is counted against your context. Billing is in DA per token, same as all other usage.

Context Window#

Search results are injected into the model's context. On long threads, be mindful that accumulated search results consume context budget. Use the max_tokens parameter if you need to cap output length.