AI · SEMICONDUCTOR · FACT CHECK
SKD WIRETrue
Liquid AI has publicly released its new flagship model, LFM2.5-24B-A2B, positioning it for on-device AI agent applications. The model and its accompanying materials are available on the company's Hugging Face repository.
Liquid AI, the AI startup known for its Liquid Foundation Models (LFM) series, has made LFM2.5-24B-A2B available through its official Hugging Face page. The release includes benchmark results and documentation on the model's training recipe and edge-inference performance, published as figures alongside the model weights.
The company is positioning the new flagship model as aimed at on-device agents — AI systems that run directly on end-user hardware rather than in the cloud, a growing focus for edge AI deployments.
The model is hosted on Hugging Face under LiquidAI's account, with benchmark, training recipe, and edge-inference figures attached to the repository.
Liquid AI's flagship release underscores the company's continuing focus on efficient models designed for local execution on consumer and edge devices, and the claim that the model targets on-device agents is true.
