What does Qwen3 Reranker 8B cost on RunInfra?

Input costs $0.05 per 1M input tokens.

What are the limits for Qwen3 Reranker 8B?

Documents per request: Up to 500. Maximum tokens per query-document pair: 16,000. Pair scoring: Each query-document pair receives a relevance score. Query length: Up to 8,192 tokens. Scored tokens per document: 4,096 by default. Longer documents are truncated for scoring and the response says so; set max_tokens_per_doc up to the pair limit to score full documents. Tokens per request: About 150,000 across all pairs, whichever of the per-request limits is reached first. Metering: Input-token metering counts the query once per document. Maximum request size: 3.5 MB per request.

RunInfraby RightNow

Dashboard Sign in Get started

RunInfraby RightNow

Join the community System status

Backed by

Combinator

AICPA Type II

SOC 2

Ask AI about RunInfra

Part of RightNow

Qwen3 Reranker 8B

qwen3-reranker-8b

Get API key View docs

Qwen3 Reranker 8B is a reranker listed in RunInfra Model APIs. RunInfra serves it as qwen3-reranker-8b at $0.05 per 1M input tokens. The API provides RunInfra native rerank, POST /v1/rerank.

Pricing

USD, pay per token

per 1M input tokens: $0.05

Access

Confirm how your client reaches this model.

Provider: Qwen
API compatibility: RunInfra native rerank, POST /v1/rerank
Accepted input: Text only
Availability: Available
Data retention: Your prompts are never stored and never used for training.

Capacity

Check the limits your workload must fit.

Documents per request: Up to 500
Maximum tokens per query-document pair: 16,000
Pair scoring: Each query-document pair receives a relevance score.
Query length: Up to 8,192 tokens
Scored tokens per document: 4,096 by default. Longer documents are truncated for scoring and the response says so; set max_tokens_per_doc up to the pair limit to score full documents.

Rerank economics

Plan candidate count against the measured p50 latency model.

Default candidates: 15 candidates. More candidates means better recall and proportionally more latency. Scores are sigmoid relevances in 0 to 1, comparable within one query; choose cutoff thresholds per corpus rather than reusing them across models.
Fixed p50 latency per search: 66 ms per search. Measured by RunInfra on one H100 PCIe with unique inputs, 264 tokens per query-document pair, and concurrency 1.
Marginal p50 latency per candidate: 8.5 ms per candidate. Measured by RunInfra on one H100 PCIe with unique inputs, 264 tokens per query-document pair, and concurrency 1.
Candidates in a 200 ms budget: About 15 candidates in a 200 ms budget. Estimated from RunInfra's measured p50 model on one H100 PCIe with unique inputs, 264 tokens per query-document pair, and concurrency 1.

Capabilities

See which request modes the API supports.

Precision: BF16, unquantized

View Full Spec

API route: Send requests to POST /v1/rerank through api.runinfra.ai.
Upstream model

Code examples

Set RUNINFRA_GATEWAY_KEY to your workspace API key before using an example.

Trust and provenance

Verify the company and operating credentials behind this API.

RunInfra is a sub-product of RightNow Research Lab.

SOC 2 Type IIAudited access, logging, and incident response.
Y CombinatorBacked by Y Combinator.
NVIDIA InceptionMember of NVIDIA Inception.

RunInfraby RightNow

Join the community System status

Backed by

Combinator

AICPA Type II

SOC 2

Ask AI about RunInfra

Part of RightNow

RunInfraby RightNow

Join the community System status

Backed by

Combinator

AICPA Type II

SOC 2

Ask AI about RunInfra

Part of RightNow

Qwen3 Reranker 8B

qwen3-reranker-8b

Get API key View docs

Qwen3 Reranker 8B is a reranker listed in RunInfra Model APIs. RunInfra serves it as qwen3-reranker-8b at $0.05 per 1M input tokens. The API provides RunInfra native rerank, POST /v1/rerank.

Pricing

USD, pay per token

per 1M input tokens: $0.05

Access

Confirm how your client reaches this model.

Provider: Qwen
API compatibility: RunInfra native rerank, POST /v1/rerank
Accepted input: Text only
Availability: Available
Data retention: Your prompts are never stored and never used for training.

Capacity

Check the limits your workload must fit.

Documents per request: Up to 500
Maximum tokens per query-document pair: 16,000
Pair scoring: Each query-document pair receives a relevance score.
Query length: Up to 8,192 tokens
Scored tokens per document: 4,096 by default. Longer documents are truncated for scoring and the response says so; set max_tokens_per_doc up to the pair limit to score full documents.

Rerank economics

Plan candidate count against the measured p50 latency model.

Default candidates: 15 candidates. More candidates means better recall and proportionally more latency. Scores are sigmoid relevances in 0 to 1, comparable within one query; choose cutoff thresholds per corpus rather than reusing them across models.
Fixed p50 latency per search: 66 ms per search. Measured by RunInfra on one H100 PCIe with unique inputs, 264 tokens per query-document pair, and concurrency 1.
Marginal p50 latency per candidate: 8.5 ms per candidate. Measured by RunInfra on one H100 PCIe with unique inputs, 264 tokens per query-document pair, and concurrency 1.
Candidates in a 200 ms budget: About 15 candidates in a 200 ms budget. Estimated from RunInfra's measured p50 model on one H100 PCIe with unique inputs, 264 tokens per query-document pair, and concurrency 1.