Liquid AI Launches LFM2.5-2.6B: On-Device Agentic Model
JosephJacks_ · x · 2026-08-06
Liquid AI has released LFM2.5-2.6B, a new open-weight model designed to run entirely on-device across phones, laptops, PCs, and robots. It can plan, call tools, and work through multi-step tasks while keeping data completely local, reducing the marginal cost of each run to essentially zero.
According to the announcement, the model was pre-trained on 34T tokens and supports a 128K context length. It delivers comparable or better scores than models nearly 4x its size, notably scoring 77.83 on the ToolSandbox benchmark, ahead of Qwen3.5-9B at 76.44.
Related event: Liquid AI Releases 2.6B On-Device Agent Model LFM2.5(14 posts)→
More from coding & agent
- A Vision for Cargo: Rust's Package Manager Eyes Agentic Development — charliermarsh · 2026-08-06
- Open-Source Benchmarks: RTX 5090 LLM Quants and 8GB VRAM Agentic Scores — max_paperclips · 2026-08-06
- Agent Harnesses Swing Accuracy by 15%: DataSpace Benchmark Reveals Shortcomings — omarsar0 · 2026-08-06
- Meta Launches Muse Code Terminal Coding Agent and Muse Spark 1.2 Model — alexandr_wang · 2026-08-06
- Christoph Nakazawa's Retrospective: Building a Framework with LLMs — cnakazawa · 2026-08-06
- Scale Details Muse Code: Co-training Model and Agent for Better Tool Use — alexandr_wang · 2026-08-06