Answer.AI and LightOn release ModernBERT encoder models

Answer.AI and LightOn released ModernBERT, with 149M and 395M encoder checkpoints and an 8,192-token sequence length. The Apache 2.0 weights support adaptation for retrieval, classification and other language-understanding tasks.

Encoders for understanding inputs

ModernBERT is an encoder family for tasks such as retrieval and classification. The checkpoints require downstream adaptation for many applications. Public weights and Apache 2.0 licensing enabled reuse; the developers' performance comparisons remain their own evaluations. [1][3]

Sources

  1. Finally, a Replacement for BERT: Introducing ModernBERT

    Answer.AI and LightOn · Dec 19, 2024

  2. ModernBERT: Better, Faster, Stronger Knowledge Retrieval and Classification

    LightOn · Dec 19, 2024

  3. ModernBERT-base

    Answer.AI and LightOn · retrieved Oct 23, 2026

  4. LightOn and Answer.AI Releases ModernBERT: A New Model Series that is a Pareto Improvement Over BERT with Both Speed and Accuracy

    MarkTechPost · Dec 20, 2024