Deploy local agents everywhere with LFM2.5-2.6B
- Source
- huggingface.co
- Author
- Liquid AI
- Date

Context
Liquid AI trained LFM2.5-2.6B to practice the multistep work expected of a local . During reinforcement learning, the model proposes actions inside agent software, tools execute them in task environments, and the resulting trajectories and rewards feed into training. The team's account describes this as a way to improve compatibility with different tools and instructions.
Its reported comparison shows especially strong instruction following among the tested models, while results vary across coding and agent tasks. The useful proposition is a small model trained for tool-driven workflows on local hardware, with suitability determined by the particular task.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
- tool use — A model's ability to call external functions — run code, search the web, edit files — instead of only generating text.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
“It supports tool calling and multi-step workflows while staying small and fast enough for everyday hardware, from laptops to phones.”
“LFM2.5-2.6B is pre-trained on ~34T tokens, with a mid-training phase that extends the context window to 128K.”
“It is the smallest model in the group, yet it competes with and often beats the rest.”
“Coding is the one place the larger models keep a clear lead, so reach for something bigger there.”
Checking sign-in…
Loading comments…






