Liquid AI's smallest model yet LFM2.5-230M beats models 4X its size at data extraction, can run 'anywhere'

2026-06-25 · Source: VentureBeat · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Data Science & Analytics · Depth: Intermediate, short

Summary

Liquid AI, founded by former MIT computer scientists, released its LFM2.5-230M language model on June 25, 2026. This 230-million-parameter foundation model is explicitly designed for on-device agentic workflows and local deployment on smartphones, laptops, and robotics. It reportedly outperforms models over 4X its size, such as the 800-million-parameter Alibaba Qwen3.5-0.8B and 1-billion-parameter Google Gemma 3 1B, on data extraction and tool-use benchmarks like BFCLv3 and CaseReportBench. The LFM2.5-230M utilizes a unique LFM2 hybrid architecture, combining gated short-range convolutions with grouped-query attention, enabling a 32K context window and maintaining a memory footprint under 400MB. It achieves decode speeds of 213 tokens per second on a Samsung Galaxy S25 Ultra and 42 tokens per second on a Raspberry Pi 5. The model is available under a dual-use commercial license, free for entities with under \$10 million in annual revenue, requiring a paid agreement for larger enterprises.

Key takeaway

For AI Engineers or Directors of AI/ML evaluating on-device AI solutions, Liquid AI's LFM2.5-230M presents a compelling option. If your team needs to automate data extraction or deploy agentic workflows on edge hardware, this 230-million-parameter model offers superior performance for its size, significantly reducing cloud compute costs and latency. You should assess its capabilities for local deployment on smartphones, robotics, or other constrained environments to streamline operations.

Key insights

Liquid AI's LFM2.5-230M demonstrates that highly efficient, small models can surpass larger ones for specific on-device data extraction and tool-use tasks.

Principles

• Architectural efficiency is key for edge AI performance.
• On-device deployment reduces cloud costs and latency.
• Specialized small models optimize targeted workflows.

Method

The LFM2.5-230M model employs a hybrid LFM2 architecture, interleaving gated short-range convolutions with grouped-query attention to process information efficiently with a 32K context window.

In practice

• Implement lightweight data extraction pipelines.
• Enable autonomous edge systems on constrained hardware.
• Structure diverse unstructured data into JSON formats.

Topics

• Liquid AI
• LFM2.5-230M
• Edge AI
• Data Extraction
• Agentic Workflows
• Small Language Models

Best for: AI Architect, NLP Engineer, CTO, AI Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• Liquid AI Releases LFM2.5-8B-A1B: An On-Device MoE Model With 8.3B Total and 1.5B Active Parameters — LFM2.5-8B-A1B is an on-device MoE model that significantly reduces hallucinations and enables complex tool-calling.
• How small can LLMs really be? — Advances in architecture and optimization enable surprisingly small, useful LLMs, expanding deployment to older and specialized hardware.
• 😸 Diffusion models are coming for text at ~$0.80 per MILLION flat — Diffusion models can significantly accelerate text processing by editing full drafts in parallel, rather than generating word-by-word.
• Gemma 4 Is INCREDIBLE! Google's Open Model IS POWERFUL! (Fully Tested) — Gemma 4 models prioritize efficiency and agentic capabilities, enabling high performance on local and edge devices.
• Latest open artifacts (#18): Arcee's 400B MoE, LiquidAI's underrated 1B model, new Kimi, and anticipation of a busy month — Smaller open models are achieving competitive performance against larger counterparts through extensive pretraining and architectural innovations.
• 🗞️ Anthropic releases Claude Opus 4.8 on the same day as its $965B valuation round. — The AI frontier is rapidly expanding with new model capabilities, inference optimizations, and valuation milestones.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by VentureBeat.