Liquid AI Releases LFM2.5-2.6B: An On-Device Agentic Model With 128K Context, Tool Calling, And Open Weights

- Liquid AI dropped LFM2.5-2.6B, an on-device agentic model that actually works without bleeding your data to a cloud API. 2.69B params, 128K context, and it runs locally on anything from an M5 Max to a robot arm. No third-party calls means no regulatory snooping and near-zero marginal cost per run. It beats larger models on tool use and instruction following, though coding is still a mess. Perfect for air-gapped finance or privacy-paranoid meat wallets who want agents that don’t sell their soul. Open weights in GGUF/MLX/ONNX mean you can self-host before the Chrome Syndicate audits your hash manifest. Finally, AI that doesn’t require a subscription to exist.