ModernBERT
Model ● Covered in 5 stories + Follow
ModernBERT is a transformer-based encoder architecture that has become a foundation for multiple downstream applications in information retrieval and embedding tasks. Recent developments include IBM building multilingual embedding models on ModernBERT supporting 200+ languages and 32K context windows, the release of reranker models based on ModernBERT encoders for retrieve-then-rerank pipelines, and successful finetuning demonstrations showing domain-specific ModernBERT models outperforming larger general-purpose alternatives. The model has also been evaluated for inference cost efficiency at scale, with benchmark comparisons showing its computational characteristics relative to other transformer-based approaches.
Updated 6 August 2026
Specifications
No specifications recorded yet.
Latest developments
2026
- Introducing the Ettin Reranker Family
- Granite Embedding Multilingual R2: Open Apache 2.0 Multilingual Embeddings with 32K Context — Best Sub-100M Retrieval Quality
- Training mRNA Language Models Across 25 Species for $165
2025
Relationships
Products & technology
- Ettin Reranker Family derived from this model · 1 source
- Granite Embedding Multilingual R2 97M derived from this model · 1 source