LFM2.5 2.6B is a 2.6 billion parameter causal language model developed by LiquidAI, notable for its efficient architecture and 128,000 token context window. It is part of the LFM2 model series and is designed for general-purpose text generation and understanding tasks. The model is publicly available for download and use on the Hugging Face platform.
What it is
The LFM2.5 2.6B model is a transformer-based causal language model, meaning it predicts the next token in a sequence autoregressively. It belongs to the LFM2 architecture family, which is designed by LiquidAI to offer a balance of performance and efficiency. With 2.6 billion parameters, it is a mid-sized model that aims to provide strong capabilities while remaining accessible for a wide range of hardware setups.
A key feature of this model is its support for a context window of 128,000 tokens, allowing it to process and generate much longer documents than many standard language models. This makes it suitable for applications requiring long-form text understanding, such as document summarization, code generation, and complex dialogue systems. The model is distributed with weights available for direct use and fine-tuning.
Key facts
| Attribute | Value |
|———–|——-|
| Released | 2026 |
| Developer | LiquidAI |
| License | Other (custom) |
| Type / Architecture | LFM2, Causal Language Model |
| Availability | Hugging Face (LiquidAI/LFM2.5-2.6B) |
How it compares
LFM2.5 2.6B occupies a space between smaller, more efficient models (like those under 1B parameters) and larger, more capable but resource-intensive models (like those over 7B parameters). Its 128K context window is a significant advantage over many models in its parameter class, which often max out at 4K to 32K tokens. It is distinct from its sibling model, LFM2.5-VL-3B, which is a 3 billion parameter vision-language model designed for multimodal tasks, whereas LFM2.5 2.6B is a text-only language model.
FAQ
What is the license for LFM2.5 2.6B?
The model is released under a custom “Other” license on Hugging Face. Users should review the specific license terms provided by LiquidAI on the model’s repository page to understand permitted uses, restrictions, and any attribution requirements before downloading or using the model.
What tasks is LFM2.5 2.6B best suited for?
The model is designed for general causal language modeling tasks, including text generation, summarization, and question answering. Its large 128K context window makes it particularly well-suited for applications that require processing long documents or maintaining coherence over extended conversations, such as long-form writing assistants or code generation with large codebases.
How do I access and run the LFM2.5 2.6B model?
The model is available for direct download and use through the Hugging Face Transformers library. Its weights are publicly hosted on the LiquidAI/LFM2.5-2.6B repository page. Users can load the model using its identifier and integrate it into their applications using standard Hugging Face pipelines for text generation.