NVIDIA NIM Reranking API

openapi: 3.1.0
info:
  title: NVIDIA NIM Reranking API
  description: >
    NVIDIA NeMo Retriever cross-encoder reranking endpoint. Given a query and a
    list of candidate passages, returns each passage's relevance score so RAG
    pipelines can re-order retrieved chunks before they hit the LLM context.
  version: '2026-05-25'
  contact:
    name: NVIDIA Developer Support
    url: https://forums.developer.nvidia.com/c/ai-data-science/nemo-llm-service/
  license:
    name: NVIDIA AI Enterprise License
    url: https://www.nvidia.com/en-us/data-center/products/ai-enterprise/
servers:
  - url: https://integrate.api.nvidia.com
    description: NVIDIA-hosted NIM endpoint
  - url: http://localhost:8000
    description: Self-hosted NIM container default
security:
  - BearerAuth: []
tags:
  - name: Reranking
    description: Cross-encoder reranking operations
paths:
  /v1/ranking:
    post:
      summary: Rank Candidate Passages
      description: Score candidate passages against a query using a NeMo Retriever reranker.
      operationId: rankPassages
      tags:
        - Reranking
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/RankingRequest'
      responses:
        '200':
          description: Ranked passages with relevance scores.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/RankingResponse'
        '400':
          description: Invalid request.
        '401':
          description: Missing or invalid API key.
        '429':
          description: Rate limit exceeded.
components:
  securitySchemes:
    BearerAuth:
      type: http
      scheme: bearer
      bearerFormat: nvapi-...
  schemas:
    RankingRequest:
      type: object
      required: [model, query, passages]
      properties:
        model:
          type: string
          description: e.g. `nvidia/llama-3.2-nv-rerankqa-1b-v2`, `nvidia/nv-rerankqa-mistral-4b-v3`.
        query:
          type: object
          required: [text]
          properties:
            text:
              type: string
        passages:
          type: array
          items:
            type: object
            required: [text]
            properties:
              text:
                type: string
        truncate:
          type: string
          enum: [NONE, END]
          default: END
    RankingResponse:
      type: object
      properties:
        rankings:
          type: array
          items:
            type: object
            properties:
              index:
                type: integer
                description: Original index in the request `passages` array.
              logit:
                type: number
                description: Raw cross-encoder relevance logit (higher = more relevant).