Liquid AI has released LFM2.5-2.6B, a 2.69‑billion‑parameter on-device agentic model with a 131K-token context window, tool calling capabilities, open weights, and two checkpoints tailored for base training and agentic workloads. Pre-trained on roughly 34 trillion tokens and covering 16 languages, the text-only model ships in GGUF, MLX, ONNX, and native formats, runs on phones to H100 GPUs, and is licensed under lfm1.0 for broad deployment. With competitive tool-use and instruction-following benchmarks against larger models, it targets automotive, consumer electronics, robotics, healthcare, finance, e-commerce, defense, and regulated or air-gapped environments needing local inference, fine-tuning options, and low marginal operating costs.
This update represents a notable development in the Ai sector. Organizations and founders tracking this space should evaluate potential strategic and technical implications on their operations.