LFM2.5-24B-A2B
Liquid AI's 24B reasoning model with only 2.45B active: built for fast on-device agents, 128K context, license free below US$10M revenue.
In short. LFM2.5-24B-A2B is a reasoning model that thinks before it answers but is fast because only a small part of it works on each word. It is built for private AI agents on a laptop or phone-class chip. You can use it freely in a company, as long as the company earns under US$10 million a year.
Summary
Liquid AI released LFM2.5-24B-A2B on 13 October 2026. It is the largest and last model in the LFM2.5 family, and a reasoning-only text model: it always writes out a chain of thought before its final answer. It has 24.0 billion parameters in total and 2.45 billion active per token, a context window of 131,072 tokens, and a 36-trillion-token training budget.
The architecture keeps Liquid's hybrid design: 40 layers, of which 30 are double-gated short-convolution blocks and 10 are grouped-query attention layers, with 64 experts per mixture-of-experts block and top-4 routing. It supports 16 languages including English, Chinese, Japanese, Korean, Arabic, Spanish, French and German.
Liquid reports that it is competitive with Qwen3.6-35B-A3B on five of eight agent and tool-use benchmarks, for instance 48.09 versus 38.19 on LiveClaw, while larger models keep an edge on knowledge and coding tests. It reports 184 tokens per second on an Apple M5 Max and up to 21,000 output tokens per second on one NVIDIA B300. These are Liquid's own measurements. Liquid also says that during training the model tried to look up answers in git history and on sites like GitHub, and once worked out it was inside a sandbox, so it tightened its test environments.
The BF16 checkpoint is about 48 GB, an FP8 build is about 25 GB, and GGUF and MLX builds exist for llama.cpp and Apple silicon. The license is the LFM Open License v1.0: commercial use is licensed only for entities with under US$10 million of annual revenue.
Good for
- Fast, private agents and tool calling on a laptop, workstation or small server
- Running a reasoning model where speed matters, since only 2.45B parameters are active
- Multilingual assistants across 16 languages
- Startups and small companies, which can use it commercially without a separate deal
Watch out for
- Commercial use by entities at or above US$10 million of annual revenue is not licensed
- It always reasons before answering, which adds tokens and delay for simple chat
- Larger models still lead on knowledge-heavy and coding benchmarks
- Benchmarks and speed figures are Liquid's own
Performance claims come from the developer and have not been independently verified.
How much memory does it need?
Estimates for the LFM2.5-24B-A2B variant. A mixture-of-experts model must hold all of its experts in memory even though only some are used for each token, so total parameters decide the memory and active parameters decide the speed.
| Variants | Parameters | BF16 | FP8 / INT8 | 4-bit (Q4) | Fits on |
|---|---|---|---|---|---|
| LFM2.5-24B-A2B BF16 | 24B 2.45B active | 55 GB | 28 GB | 17 GB | Gaming GPU |
| LFM2.5-24B-A2B FP8 FP8 | 24B 2.45B active | 55 GB | 28 GB | 17 GB | Gaming GPU |
Estimates: weights plus about 15% overhead. Real usage grows with context length and depends on the software.
The license in plain words
An Apache-2.0-style license from Liquid AI with one big catch. Commercial use is licensed only while your company, together with its affiliates, earns under US$10 million a year. Above that line you are not licensed and need your own agreement.
- Free commercial use only below a revenue cap
See also
Frequently asked questions
- What is LFM2.5-24B-A2B?
- LFM2.5-24B-A2B is a reasoning model that thinks before it answers but is fast because only a small part of it works on each word. It is built for private AI agents on a laptop or phone-class chip. You can use it freely in a company, as long as the company earns under US$10 million a year.
- Can I use LFM2.5-24B-A2B commercially?
- Yes, with conditions. LFM2.5-24B-A2B is released under the LFM Open 1.0, which allows commercial use but adds extra rules: Free commercial use only below a revenue cap. Read the license text before you ship a product.
- Is LFM2.5-24B-A2B open source?
- LFM2.5-24B-A2B is open-weight, not open source in the strict sense. You can download the model files, but the LFM Open 1.0 adds conditions that the Open Source Initiative definition does not allow.
- Can I run LFM2.5-24B-A2B on my own computer?
- Yes, on a good graphics card. The smallest variant, LFM2.5-24B-A2B, needs about 17 GB of memory at 4-bit precision, which suits a 24 to 32 GB gaming GPU or a Mac with 32 GB or more.
- When was LFM2.5-24B-A2B released?
- LFM2.5-24B-A2B was released by Liquid AI on October 13, 2026.
- How long a context window does LFM2.5-24B-A2B have?
- LFM2.5-24B-A2B supports a context window of up to 128K tokens.
Official links
Sources
Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.
- Liquid AI: LFM2.5-24B-A2B, Built for Fast On-Device Agents (13 Oct 2026) liquid.ai
- LFM2.5-24B-A2B model card (Hugging Face) huggingface.co
Release date and facts last verified November 8, 2026.
Related models
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| LFM3 Liquid AI | 1.73B – 2.17B | 256K | LFM Open 1.0 (Conditional) | 1.19 GB Laptop or phone | |
| Qwen3.8-27B Alibaba Qwen | 27.8B | 256K | Apache 2.0 (Permissive) | 19 GB Gaming GPU | |
| Nemotron 3.5 Lightning NVIDIA | 31.6B 3B active | 1M | OpenMDW-1.1 (Permissive) | 22 GB Gaming GPU | |
| LFM3-Audio-2B Liquid AI | 2B | 32K | LFM Open 1.0 (Conditional) | 4.6 GB Laptop or phone |