Reranking - Planck

POST /v1/rerank scores and reorders candidate documents against a query. It is commonly used after vector or keyword retrieval to improve the final context sent to a generation model.

Rerank Documents

curl https://api.inquantum.ai/v1/rerank \
  -H "Authorization: Bearer $PLANCK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "fireworks/qwen3-reranker-8b",
    "query": "How do I reset my API key?",
    "documents": [
      "Open the API Keys page and select Rotate key.",
      "Invoices are available from the Billing page.",
      "You can change your display name in Profile settings."
    ],
    "top_n": 2,
    "return_documents": true
  }'

A successful response contains normalized results, ordered from highest to lowest relevance. Every result includes its original document index and a relevance_score.

Choosing a Model

Copy the exact rerank model ID from the model registry, including every /-separated part. You always call the same Planck endpoint with your Planck API key. Some rerank models may require BYOK when their billing usage cannot be verified safely.

Reranking is not part of the OpenAI or Anthropic chat SDK surface. Call the JSON endpoint directly with fetch, your HTTP client, or cURL.

Billing and Tracking

Rerank requests appear in Requests with Planck-Operation: rerank, the resolved provider/model identity, duration, billing data when available, and the request and normalized response bodies. See the Rerank API reference for supported fields and response shape.