LFM2.5-24B-A2B

LFM2.5-24B-A2B is a language model that can plan tasks and call tools on a user’s own hardware. Liquid AI released it on 13 October 2026, with downloadable base and instruction-trained versions.

It combines convolutions with attention

The model has 24 billion parameters and selects a smaller set of experts for each token. Its hybrid design combines short convolutions with grouped-query attention. The model card specifies a 131,072-token context window. The A2B name describes the active-parameter size in rounded form.

Local deployment keeps model inference on the selected machine. Any tools attached to an agent still have their own network access and data-handling behaviour. Runtime, memory, quantisation and task length affect the experience.

Current hosted prices and licence conditions

Liquid AI’s pricing page, checked on 15 October, lists $0.05 per million input tokens and $0.70 per million output tokens. These are the current published prices; the launch announcement carried earlier rates.

The downloadable model uses Liquid’s own licence. Its published terms allow production use without a licence payment for businesses below $10m in annual revenue; larger businesses should check the enterprise terms. Research, education and non-profit use have separate provisions. Read the full licence before redistribution or commercial deployment.

See the Liquid model-family guide for the wider line.