Voyage AI has released Rerank-2, their next-generation reranking model that establishes new benchmarks across multiple retrieval tasks.
Key Improvements
The new model offers significant improvements over its predecessor:
- 15% better MRR on standard retrieval benchmarks
- Improved latency with optimized inference
- Better handling of long documents up to 8K tokens
- Domain-specific variants for legal and technical content
What This Means for RAG Systems
Reranking is a critical component of production RAG systems. Better reranking directly translates to more relevant responses and fewer retrieval failures.
At SmartFAQs.ai, we've integrated Voyage Rerank-2 into our pipeline, and early results show measurable improvements in answer quality.
Getting Started
Voyage Rerank-2 is available today through the Voyage AI API. Existing customers can upgrade by changing their model parameter.