LFM2.5-24B-A2B is an open-weight Mixture-of-Experts reasoning model developed by Liquid AI for fast, private agents running on-device.
The model always produces an explicit reasoning trace before its final answer and is post-trained for tool calling and multi-step agentic tasks.
| Developer | LiquidAI |
|---|---|
| Model ID | liquid/lfm-2.5-24b-a2b |
| Release Date | October 13, 2026 |
| Context Window | 128,000 tokens (~128K) |
| Max Output Tokens | 8,192 tokens |
| Modalities | Text → Text |
| Capabilities | reasoning, streaming, structured (JSON schema) output, tool/function calling |
Sparse MoE design
The model has 24B total parameters with 2B active parameters per token.
Explicit reasoning
LFM2.5-24B-A2B always thinks before answering, using a <think> tag in its chat template.
Tool calling
The model supports tool calling and can interpret tool outputs before producing a final answer.
Open weights
The base and post-trained model versions are open-weight and available through Hugging Face and Liquid AI's Playground.
On-device agents
Its sparse architecture and on-device focus make it suitable for private agents running on local hardware.
Multi-step automation
The model is post-trained for tool calling and multi-step agentic tasks.
Structured tool workflows
It can generate function calls, execute tool-assisted workflows, and return a final response based on tool output.
Sources: https://www.liquid.ai/blog/lfm2-5-24b-a2b · https://docs.liquid.ai/lfm/models/lfm25-24b-a2b · https://huggingface.co/LiquidAI/LFM2.5-24B-A2B
Fast, private on-device agents that reason through multi-step tasks and call tools reliably, including agents running continuously on local hardware.
Knowledge-heavy applications that combine the model with retrieval or web search.
Local document-processing workflows, such as extracting text, comparing documents, and generating PDFs.
Security and compliance workflows, including scanning files for leaked credentials or personal data and producing audit trails for tool calls.
import requests r = requests.post( "https://api.aimlapi.com/v1/chat/completions", headers={"Authorization": "Bearer " + AIMLAPI_KEY}, json={ "model": "liquid/lfm-2.5-24b-a2b", "messages": [ { "role": "user", "content": "Hello!" } ] }, ) print(r.json())
const r = await fetch("https://api.aimlapi.com/v1/chat/completions", { method: "POST", headers: { Authorization: `Bearer ${process.env.AIMLAPI_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ "model": "liquid/lfm-2.5-24b-a2b", "messages": [ { "role": "user", "content": "Hello!" } ] }), }); console.log(await r.json());
curl -X POST https://api.aimlapi.com/v1/chat/completions \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"liquid/lfm-2.5-24b-a2b","messages":[{"role":"user","content":"Hello!"}]}'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Input | |
| Output | |
| Cached input |
Liquid Console offers $2 in signup credit, and cached input is billed at a lower rate than uncached input.
| Benchmark | Score | What it measures | Source | Retrieved |
|---|
The primary published source is Liquid AI’s official announcement, “LFM2.5-24B-A2B: Built for Fast On-Device Agents.”[1 Source: https://www.liquid.ai/blog/lfm2-5-24b-a2b
