reranker · live · batch-class

Qwen3-Reranker-8B

Specifically designed for text embedding and ranking tasks. Building upon the dense foundational models of the Qwen3 series... the reranking model excels in various text retrieval scenarios.

Qwen’s own description of the model, from Qwen3-Reranker-8B model card. Served here as qwen/qwen3-reranker-8b.

What Qwen built it for

in the vendor’s words

“the reranking model excels in various text retrieval scenarios.”
“both embedding and reranking models support user-defined instructions to enhance performance for specific tasks, languages, or scenarios.”
“The Qwen3 Embedding series offer support for over 100 languages.”

Quoted from Qwen3-Reranker-8B model card, retrieved 2026-08-27.

Recommended usage

what Qwen recommends

Instructions

“using instructions (instruct) typically yields an improvement of 1% to 5% compared to not using them.”

On this endpoint: pass your task instruction in the request’s instruction field; omitted, the vendor’s default web-search instruction is used.

Scoring

Each (query, document) pair is judged with Qwen’s own prompt and scored as the probability the model answers “yes” — a calibrated 0-to-1 relevance per document, returned sorted. The judge sees up to the full 33K-token window per pair.

Context length

Qwen gives the window as 33K tokens per query+document pair, which is what we serve — long documents go in whole instead of pre-truncated.

What to point it at

Qwen names these capabilities for the series:

“text retrieval, code retrieval, text classification, text clustering, and bitext mining”
“support for over 100 languages”, including programming languages

Recommendations on this page are Qwen’s, from Qwen3-Reranker-8B model card, retrieved 2026-08-27.

What it does on this endpoint

our gates · not vendor figures

exactness score + order parity vs the fp16 reference, on the deployed artifact

scores 0–1, sorted P("yes") per pair, deterministic

context 33K tokens per query+document pair

No speed figure is published for this model on purpose: throughput is best-effort by contract (see the price section). What IS gated, on every deploy, is byte-level correctness — the served scores are checked against the fp16 reference for score and rank-order parity on the exact artifact and binary serving, before cutover.

Price, and the scheduling contract behind it

per million tokens · input only · no request fee

Token rates — Qwen3-Reranker-8B
Input	Output
$0.03	$0.00 — a rerank request generates nothing

This is a batch-class surface by design: rerank traffic rides the serving valleys of the generation models and is shed with a retryable 429 when their lanes are busy. The paying interactive lanes are never taxed — and that subordinate scheduling is exactly what this price buys. Retry with backoff; throughput is best-effort, correctness is gated.

One price, no tiers, prepaid credit — the same balance as every other model here. Full price sheet.

Capabilities

qwen/qwen3-reranker-8b

Cohere-shaped wire format

A query and a documents array in, relevance-sorted scores out, with top_n and return_documents — the shape existing rerank clients already speak.

Same key, same balance

One base URL, https://api.tiyuvta.ai/v1; the model field routes. A RAG stack runs retrieval, rerank and generation on a single prepaid credit — no second vendor, no second bill.

Deterministic output

The same (query, document, instruction) triple returns the same score, byte-for-byte — safe to cache on your side and safe to compare across runs.

Pairs with

Qwen3-Embedding-8B — embed with it, rerank with this: the complete retrieval stack from one vendor family, on one key.

Where it fits best

use it for

The RAG quality lever: rerank the top 20-100 hybrid-retrieval hits before the LLM sees them
Pairs with the embedder for a complete retrieval stack on one API key
Long documents: 32k-token judge window per query-document pair

Generating over what you retrieve? Ornith-1.5 35B-A3B and Qwen3.8 27B — same key, same balance.

First request

qwen/qwen3-reranker-8b

shell — rerank https://api.tiyuvta.ai/v1

curl https://api.tiyuvta.ai/v1/rerank \
  -H "Authorization: Bearer $TIYUVTA_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "qwen/qwen3-reranker-8b",
       "query": "how do I rotate an API key?",
       "documents": ["...", "...", "..."],
       "top_n": 3}'

Scores come back sorted; top_n trims the response to the best hits.

shell — the retry loop batch-class

# a 429 means the interactive lanes are busy — retry with backoff
for i in 1 2 3 4 5; do
  code=$(curl -s -o out.json -w "%{http_code}" https://api.tiyuvta.ai/v1/rerank \
    -H "Authorization: Bearer $TIYUVTA_KEY" \
    -H "Content-Type: application/json" -d @request.json)
  [ "$code" != 429 ] && break
  sleep $((i * i))
done

A 429 is the batch-class contract working, not an outage: the interactive lanes were busy and your request should come back with backoff.