Jina Reranker v3.5
Jina Reranker v3.5 is a 0.6-billion-parameter multilingual model that scores several candidate documents together. It retains v3's request interface while targeting legal, medical and structured-data retrieval.
How the model works
Its Qwen3-based design interleaves sliding-window attention with global attention. The final global layer supports Jina's last-but-not-late method: query and documents interact in one context before relevance scoring. A small projection network produces the scoring representation. The documented context is approximately 131K tokens.Jina Reranker v3.5 model details
Use and access
The API accepts the same request schema as v3 with a changed model identifier. Downloadable versions include full weights, GGUF and MLX distributions; they are packaging choices for the same release. Jina reports domain and efficiency gains under its stated evaluation protocol, not a guarantee for every collection. Local commercial use requires terms beyond the CC BY-NC 4.0 weight license.Jina Reranker v3.5 model detailsJina Reranker v3.5 technical overview
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | API, Downloadable |
| Model identifiers | jina-reranker-v3.5jinaai/jina-reranker-v3.5jina-reranker-v3.5-GGUFjina-reranker-v3.5-mlx |
| Context window | 131,072 tokens 131K-class combined query and candidate context; the model configuration uses 131,072 tokens. Jina Reranker v3.5 model details |
| Parameters | Approximately 0.6 billionJina Reranker v3.5 model details |
| Weight license | CC BY-NC 4.0Jina Reranker v3.5 model details |
| Base model | Qwen3-0.6BJina Reranker v3.5 model details |
Related models
- Jina Reranker v3
Jina AI · Jina Reranker
A compact ranker that compares a query with several documents in one pass.
Launch date API · Downloadable - Jina Reranker m0
Jina AI · Jina Reranker
A relevance scorer for text and document images in multiple languages.
Launch date API · Downloadable