Qwen3 Reranker 8B

Qwen3 Reranker 8B is an 8-billion-parameter model within the Qwen3 family designed specifically for text embedding and reranking tasks. It is built on the dense foundational models of the Qwen3 series and is part of a lineup that also includes 0.6B and 4B sizes, giving teams a range of options for balancing reranking quality against compute and latency requirements. Its placement under the Reranker category on hosted model catalogs confirms it is intended for passage and document relevance scoring rather than open-ended generation.

As a rerank-only model, Qwen3 Reranker 8B is best suited to retrieval-augmented pipelines that already have an efficient first-stage retriever and need a more accurate relevance reranker to refine top results. The larger 8B configuration offers higher reranking precision than the smaller Qwen3 siblings in exchange for more compute, making it a good fit when result quality matters more than throughput. Teams adopting it typically pair it with a separate embedding model for initial recall and reserve Qwen3 Reranker 8B for the final scoring stage, where its larger capacity can improve ordering at the top of the ranked list.

Novita AIqwen/qwen3-reranker-8bqwen

Open provider page Provider docs

Quick Info

Provider: Novita AI
Model key: qwen/qwen3-reranker-8b
Release date: Jun 5, 2025
Last updated: Jun 5, 2025
Input modalities
Output modalities
Capabilities

Cost

A provider subscription or plan supersedes token-based pricing for this model.

Limits

Output tokens: 0 tokens
Context window: 32,768 tokens

Latest news about Qwen3 Reranker 8B

Novita AI

Coverage

Novita AI announced hosting of Qwen3 Reranker 8B on its API platform, making the model available to developers for integration into retrieval pipelines. The blog describes rerankers as tools that reorder top-k retrieved documents using relevance scoring, improving precision in search and RAG workflows. According to the Novita AI post, Qwen3 Reranker 8B shows strong benchmark performance on multilingual and code-based tasks, with evaluation suites listed including MTEB-R, CMTEB-R, MMTEB-R, MLDR, MTEB-Code, and FollowIR. The page positions the model as complementary to embedding-based retrieval for filtering noise and prioritizing relevant results.

Novita AI

Coverage

The official Qwen Hugging Face model card describes Qwen3-Reranker-8B as an 8-billion parameter text reranking model with a 32k context length and support for over 100 languages. It is part of the Qwen3 Embedding series, which also includes 0.6B, 4B, and 8B embedding variants designed for text and code retrieval. According to the card, the 8B embedding sibling ranked No. 1 on the MTEB multilingual leaderboard as of June 5, 2025 with a score of 70.58, while the reranker is described as excelling in text retrieval scenarios. The page lists both embedding and reranking models as instruction-aware, enabling custom prompts for task- or language-specific tuning.