Liquid AI Releases LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: Bidirectional Encoders That Stay Fast at 8K Context on CPU
Liquid AI launched open-weight LFM2.5-Encoder-230M and 350M models featuring 8,192-token context. These bidirectional encoders utilize a hybrid backbone optimized for CPU inference.
Liquid AI has introduced two new encoder models, the LFM2.5-Encoder-230M and LFM2.5-Encoder-350M. These open-weight models are designed as bidirectional encoders built upon the company's LFM2 hybrid architecture backbone.
Both models support an 8,192-token context window. According to the release details, the 350M variant achieved a fourth-place ranking among 14 models on a combined 17-task evaluation suite covering GLUE, SuperGLUE, and multilingual benchmarks.
The emphasis on CPU compatibility suggests a focus on efficient inference without requiring specialized GPU hardware. This positioning could appeal to developers seeking cost-effective embedding solutions for retrieval-augmented generation or semantic search tasks.
Liquid AI continues to expand its LFM series, which aims to balance performance with computational efficiency. The release adds to the growing ecosystem of open-weight encoders available for enterprise and research integration.
This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.