nvidia · Reranking

Nemotron-3.5-Rerank-8B-BF16 API

Nemotron-3.5-Rerank-8B-BF16 from nvidia scores candidate documents against a query. Reranking improves the ordering of retrieved results; it neither creates embeddings nor generates a chat answer.

Available

Toman per million tokens

Input14,390

The usage rate is fixed at request admission. Reported consumption determines the final usage amount.

API model ID
di-nvidianemotron-35-rerank-8b-bf16-2752e71a5e
Model developer
nvidia
Service type
Reranking
Current availability
Available
Configured context limit
32,768 tokens

Model features

No additional gateway capabilities have been confirmed for this model.

What is Nemotron-3.5-Rerank-8B-BF16?

Use the public identifier di-nvidianemotron-35-rerank-8b-bf16-2752e71a5e for Nemotron-3.5-Rerank-8B-BF16 in ServerNet. The gateway uses this identifier to select the configured route; the developer is nvidia.

Where to use Nemotron-3.5-Rerank-8B-BF16

  • Rerank the candidates from your search index with Nemotron-3.5-Rerank-8B-BF16.
  • Place the most relevant passages first in a retrieval augmented generation workflow.
  • Compare ranking quality and added latency on representative queries before using Nemotron-3.5-Rerank-8B-BF16 for every search.

Build your integration

Send query and documents to /v1/rerank, then order candidates by their relevance scores. Retrieve a shortlist first so the reranker sees only documents worth comparing.

Inputs and service limits

Use text queries and up to 128 text documents. The query is paired with each candidate, so adding long documents increases input consumption. Media and unsupported native parameters are not forwarded.

A request cost example

For 1,000 input tokens and 0 output tokens, without a cache discount. This is a catalogue estimate, not a fixed request fee. Reported consumption and the price frozen for your request determine the usage amount.

Estimated usage, Toman
15

Connect to Nemotron-3.5-Rerank-8B-BF16 API

Enable ai:rerank on your key and ai:models:read to check availability. Set SERVERNET_API_KEY in your server environment. Change the idempotency key for each new request; keep it unchanged only for an identical retry.

curl https://servernet.cloud/v1/rerank \
  -H "Authorization: Bearer $SERVERNET_API_KEY" \
  -H 'Content-Type: application/json' \
  -H 'Idempotency-Key: CHANGE_FOR_EACH_NEW_REQUEST' \
  -d '{
    "model": "di-nvidianemotron-35-rerank-8b-bf16-2752e71a5e",
    "query": "What is the capital of Iran?",
    "documents": [
        "Tehran is the capital of Iran.",
        "Paris is the capital of France."
    ]
}'

Use the same Idempotency-Key when retrying an identical request. Successful responses are replayed without a second charge for 24 hours. Do not automatically retry when x-should-retry:false is present. The key is shown once. Store it in a server environment variable. Keep it out of browser code, GitHub and logs.

Keys, budgets and usage reports

Frequently asked questions

Is Nemotron-3.5-Rerank-8B-BF16 available through ServerNet?

Nemotron-3.5-Rerank-8B-BF16 is currently available on the configured route. Check GET /v1/models before sending a request: availability can change with provider capacity, catalogue freshness or account configuration.

Which model ID and API endpoint should I use for Nemotron-3.5-Rerank-8B-BF16?

Use di-nvidianemotron-35-rerank-8b-bf16-2752e71a5e as the model field, the endpoint shown below and the matching key scope. ServerNet keys are project-specific; an SDK also needs https://servernet.cloud/v1 as its base URL.

How is Nemotron-3.5-Rerank-8B-BF16 usage charged?

Reranking uses its verified input-token tariff and reported request cost. The query may be counted with each candidate. Relevance scores do not create output-token charges.

What should I check before using Nemotron-3.5-Rerank-8B-BF16 in production?

Use text queries and up to 128 text documents. The query is paired with each candidate, so adding long documents increases input consumption. Media and unsupported native parameters are not forwarded.

Compare models for the same task

Qwen

Qwen3-Reranker-0.6B

Available Reranking

Configured context limit: 32,768 tokens

Input2,878
Toman per million tokens
Qwen

Qwen3-Reranker-4B

Available Reranking

Configured context limit: 32,768 tokens

Input7,195
Toman per million tokens
Qwen

Qwen3-Reranker-8B

Available Reranking

Configured context limit: 32,768 tokens

Input14,390
Toman per million tokens
Only available models are exposed by the API. Availability requires a configured provider, valid prices and a supported execution path.