1.7B
- input
- $0.00
- output
- $0.00
- Context
- 512tokens
- Listed
- 2mo ago
- Size
- 350Mparameters
Cost/1M tokens
Liquid AI's LFM2.5-Embedding-350M is a text embedding model built on the LFM2.5-350M-Base checkpoint and, alongside LFM2.5-ColBERT-350M, one of the first bidirectional members of the LFM family. The maker describes it as a dense bi-encoder producing one vector per document for the smallest, fastest index and a drop-in replacement for RAG pipelines targeting fast, cheap, reliable multilingual search across 11 languages. It is built for short-context retrieval like e-commerce semantic search, support knowledge bases, on-device semantic search, and enterprise knowledge assistants.
How it compares
Size class
Tiny350MAn open-weight model under 5 billion parameters, small enough for phones, laptops and edge devices.
Open models of a similar size from other labs
Vs. your stack
Add the models you use to see how this one compares.
Sign inFrequently asked questions
Running locally
By our estimate, LFM2.5-Embedding-350M needs about 0.7 GB at Q4, 0.9 GB at Q8 and 1.3 GB at FP16 of VRAM. That covers the weights plus headroom for a short context, and longer contexts need more.
Yes, by our estimates LFM2.5-Embedding-350M fits on an 8 GB GPU at FP16 (about 1.3 GB) and at higher precision on larger cards.
Yes, a Mac with at least 8 GB of unified memory should run LFM2.5-Embedding-350M at Q4, which needs an estimated 0.7 GB. This assumes macOS gives the GPU about 75% of unified memory.
Privacy and data
Yes, LFM2.5-Embedding-350M has open weights, so you can run it on your own hardware and keep prompts on your own machines.
Capabilities and specs
LFM2.5-Embedding-350M has a 512 token context window. It can write up to 460 output tokens per response.
No, LFM2.5-Embedding-350M takes text input only.
Quality and benchmarks
Similar models from other labs include Nemotron 3 Embed 1B (NVIDIA), Nemotron 3.5 Embed 8B (NVIDIA) and Titan Multimodal Embeddings G1 (Amazon). They are the closest matches by kind and release date.
Nemotron 3 Embed 1BLFM2.5-Embedding-350M vs Nemotron 3 Embed 1BNemotron 3.5 Embed 8BLFM2.5-Embedding-350M vs Nemotron 3.5 Embed 8BTitan Multimodal Embeddings G1LFM2.5-Embedding-350M vs Titan Multimodal Embeddings G1
Versions and lineage
LFM2.5-Embedding-350M costs $0.00 input and $0.00 output per 1M tokens, versus $0.35 and $0.50 for LFM2.5-2.6B. LFM2.5-Embedding-350M has a 512 token context window, versus 128K for LFM2.5-2.6B.
Licensing and access
LFM2.5-Embedding-350M has open weights, so you can download and run it yourself.
The LFM2.5-Embedding-350M weights are on Hugging Face at LiquidAI/LFM2.5-Embedding-350M.
Basics
LFM2.5-Embedding-350M was made by Liquid AI. It was first listed on OpenRouter on August 18, 2026.
Yes, LFM2.5-Embedding-350M has open weights, so you can run it on your own hardware without API fees.
Generated from Vertical Tabs data, updated November 2, 2026.
Discussion
No comments yet
Be the first to share your thoughts.