What LFM2.5-24B-A2B is
Liquid AI released LFM2.5-24B-A2B on October 13, 2026. According to Liquid AI:
- It has 24 billion parameters in total and 2 billion active per token.
- The context is 131,072 tokens.
- On a MacBook Pro with an M5 Max, it generated up to 184 tokens per second.
- It scored ahead of larger models on agent and tool tests, such as BFCLv4 and ToolSandbox.
On knowledge and coding, the larger models scored ahead. For that, Liquid AI suggests combining the model with search.
Who it is for
- Teams that want an assistant that acts on the computer without sending the request to an external provider.
- Routines with many tool calls, such as spreadsheets, internal systems and email.
Why through Rota Nacional
- We install it on your team's computers or run it on Rota's workers, on the same API.
- By default, CPF and CNPJ numbers, email addresses, phone numbers and names leave the text before the model.
- We check the license with your organization before deployment.
How it works
You describe the case
The tasks, the tools and your team's computers.
We test with your samples
Liquid AI does not list the model's languages. So we measure the quality in Brazilian Portuguese with your own tasks.
We connect it to your access
The model joins your team's computers or the API you already use.
Good to know
- LFM Open License v1.0, checked on Hugging Face on October 17, 2026. Commercial use only applies to organizations with annual revenue below 10 million dollars, counting the companies of their group.
- Liquid AI has not published the memory it needs or its list of languages.
- On-demand deployment: the model goes live when an organization asks for it.