# Liquid AI — LFM2.5-Encoders: Fast at Long Context, Even on CPU

- Company: Liquid AI (liquid.ai)
- Announced: 2026-07-28
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.liquid.ai/blog/lfm2-5-encoders
- Record: https://forck.live/items/16777-lfm2-5-encoders-fast-at-long-context-even-on-cpu
- Subject: LFM / d1 models
- Open weights: yes
- Models affected: LFM2.5-Encoder-230M, LFM2.5-Encoder-350M
- Context window: 8,192 tokens

Liquid AI released two bidirectional encoder models, LFM2.5-Encoder-230M and LFM2.5-Encoder-350M, built on the LFM2 hybrid architecture. They are designed for classification, natural language understanding, and token-level tasks, matching the quality of larger encoders while scaling more gently with input length up to 8,192 tokens. The models are available on Hugging Face and are particularly fast on CPU, with the 230M model being 3.7x faster than ModernBERT-base at 8,192 tokens.

## Evidence

Verbatim from https://www.liquid.ai/blog/lfm2-5-encoders:

> At 8,192 tokens, ModernBERT-base takes over a minute and a half per forward pass versus about 28 seconds for LFM2.5-Encoder-230M, which is about 3.7x faster.

---

Record: https://forck.live/items/16777-lfm2-5-encoders-fast-at-long-context-even-on-cpu
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md