Sonar-deep-research taking longer than expected

Bug Reports
1

:bug: Describe the Bug

When asking for a sonar-deep-research using the async API, I poll to check for the status of the deep research for about 30-40 minutes I get a status of IN_PROGRESS. When it eventually completes the payload returns to me as only took about 2 minutes.

    "created_at": 1768872046,
    "started_at": 1768872046,
    "completed_at": 1768872180,

:white_check_mark: Expected Behavior

After 2 minutes or the completed_at time the poll endpoint should return the research as completed.

:cross_mark: Actual Behavior

Takes 40+ minutes for the retrieve result api to send back as completed, and the completed time says it only took 2 minutes

:counterclockwise_arrows_button: Steps to Reproduce

  1. Call the API with the following request:
    curl --location ‘https://api.perplexity.ai/async/chat/completions’
    –header ‘Authorization: Bearer ’
    –header ‘Content-Type: application/json’
    –data ‘{
    “request”: {
    “model”: “sonar-deep-research”,
    “messages”: [
    {
    “role”: “user”,
    “content”: “Provide an in-depth analysis of the impact of AI on global job markets over the next decade.”
    }
    ]
    }
    }’ | jq
  2. Take the id and call the polling endpoint

curl --location ‘https://api.perplexity.ai/async/chat/completions/{REPLACE_WITH_REQUEST_ID}’
–header “Authorization: Bearer ”

:pushpin: API Request & Response (if applicable)

:globe_showing_europe_africa: Environment

  • API Version: sonar-deep-research
  • SDK (if applicable): ai-sdk v6
  • Operating System: macos

:paperclip: Logs or Screenshots (if applicable)

Add any logs or screenshots that can help debug the issue.

:memo: Additional Context

Add any other context about the problem here.

2

Hi @sselepahs, I’m experiencing the same issue for the past 4 days.

When using the async API for sonar-deep-research, the job appears to complete quickly based on the completed_at timestamp, but the polling endpoint continues to return IN_PROGRESS for a long time (around 20–30 minutes). During this time, my system keeps polling the endpoint 70–80 times before it finally returns COMPLETED,and at last it results with TIMEOUT error instead of successful completion.

This is causing two major issues:

  • Unnecessary repeated polling (infrastructure overhead)

  • Tokens continuing to be counted during polling, which directly increases cost

Is there any update or acknowledgment from the Perplexity team on this issue? Are there any recommended workarounds (e.g., backoff strategy, max polling duration, or alternative endpoint) until this is resolved?