Tiny and cheap to run, it suits bulk data extraction and tool calling; larger companies need a separate commercial licence.
Kenning's take
What it does
LFM2.5-350M is a small Liquid AI language model, retrained with more data and reinforcement learning, aimed at tool use, data extraction and structured output. Liquid says it beats models more than twice its size on instruction following and tool-use benchmarks while running on anything from cloud GPUs to cheap CPUs. Base and post-trained versions are on Hugging Face, Liquid's LEAP platform and its Playground under the LFM Open License.
- 350M parameters
- Pretraining extended from 10T to 28T tokens
- Tool use, data extraction and structured outputs
- Runs on CPUs, NPUs and GPUs
- Base and post-trained checkpoints
Who it suits. Developers running large data-extraction jobs or function calling on edge devices.
Key facts
- Category
- AI models
- Made by
- Liquid AI
- Released
- 31 March 2026
- How to get it
- open weights on Hugging Face, LEAP and the Liquid Playground
- Works on
- self-hosted, Hugging Face, on-device
- In the UK
- Yes
- Free plan
- free weights for research, non-commercial use, and commercial use under $10M annual revenue
FAQ
When was LFM2.5-350M released?
Liquid AI released it on 31 March 2026.
What is LFM2.5-350M good at?
Tool use, data extraction and structured outputs, at a size that runs on cheap CPUs.
Where can I download LFM2.5-350M?
On Hugging Face, Liquid's LEAP platform and its Playground.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.