Liquid AI Launches Smallest Model LFM2.5-230M with Multi-Framework Support
Decision Brief
What changedLiquid AI releases LFM2.5-230M, supporting llama.cpp, MLX, vLLM, SGLang, and ONNX, achieving 213 tok/s on Galaxy S25 Ultra.
Why it mattersThis tiny model excels in tool usage and data extraction, outperforming larger models despite its 230M parameters, crucial for resource-constrained edge deployment.
Who should careTeams building on model APIs
Affected stackQwenLlama
Builder actionEvaluate
Source confidenceMedium · Reliable media or first-hand reporting
Liquid AI has unveiled LFM2.5-230M, a 230M-parameter open-weight model, their smallest to date. Based on the LFM2 architecture, it targets tool use and data extraction. In benchmarks, it surpasses Qwen3.5-0.8B and Gemma 3 1B in instruction following. Edge inference is impressive: 213 tok/s on Galaxy S25 Ultra and 42 tok/s on Raspberry Pi 5. The model supports llama.cpp, MLX, vLLM, SGLang, and ONNX.
Summary basis: official / RSS sourceUnless it says 'full article read', this summary is based only on publicly available content — it never pretends to have read restricted originals.
Sources
- MarkTechPost
Fast research-paper and ML tooling summaries, useful for infra and agent updates.
- MarkTechPost