Deep Research

plamo-3.1-prime-deep-research is a model that searches the web repeatedly for a question and writes a report with citations. Use it through the Responses API with background: true as an asynchronous job. A report takes from a few minutes to tens of minutes. Results are kept for 24 hours after the job finishes.

Sample Usage

Request

curl

$ curl \
    -H "Authorization: Bearer ${PLAMO_API_KEY}" \
    -H "Content-Type: application/json" \
    "https://api.platform.preferredai.jp/v1/responses" \
    -d @- << EOF
{
  "model": "plamo-3.1-prime-deep-research",
  "input": "Research recent trends in generative AI regulation in Japan",
  "background": true,
  "store": true
}
EOF

Response

right after creation

{
    "id": "resp_3f9a1c...",
    "object": "response",
    "created_at": 1791272415.0,
    "status": "queued",
    "background": true,
    "model": "plamo-3.1-prime-deep-research",
    "output": [],
    "usage": null,
    "tool_usage": null,
    "error": null,
    "tools": [{"type": "web_search"}],
    "max_tool_calls": 100,
    "incomplete_details": null
}

after completion

{
    "id": "resp_3f9a1c...",
    "object": "response",
    "status": "completed",
    "output": [
        {
            "type": "web_search_call",
            "id": "ws_...",
            "status": "completed",
            "action": {"type": "search", "query": "generative AI regulation Japan", "queries": ["generative AI regulation Japan"]}
        },
        {
            "type": "reasoning",
            "id": "rs_...",
            "summary": [{"type": "summary_text", "text": "..."}]
        },
        {
            "type": "message",
            "id": "msg_...",
            "role": "assistant",
            "status": "completed",
            "content": [
                {
                    "type": "output_text",
                    "text": "# Generative AI Regulation in Japan\n\n...",
                    "annotations": [
                        {"type": "url_citation", "start_index": 120, "end_index": 180, "url": "https://...", "title": "..."}
                    ]
                }
            ]
        }
    ],
    "usage": {"input_tokens": 182340, "output_tokens": 21530, "total_tokens": 203870},
    "tool_usage": {"web_search": {"num_requests": 18}}
}

Notes

• A job runs for at most about 25 minutes. Past that, it ends with status: "failed" and error.code: "timeout".
• If part of the research (such as writing a chapter heading) fails, that part is skipped, the report is written to the end, and the job ends with status: "incomplete". What is missing is written in the summary of a reasoning item.
• Charges are based on the tokens used to write the report and the number of searches in tool_usage.web_search.num_requests.