Liquid AI reveals 8B-A1B MoE trained on 38T
Summary
Liquid AI announces LFM2.5-8B-A1B, an 8B-A1B on-device mixture-of-experts model trained on 38T tokens with a 128K context window. The model runs edge AI on consumer hardware with day-one support for llama.cpp, MLX, vLLM, SGLang, and supports deployment via Hugging Face and Playground; benchmarks show improved accuracy and reduced hallucinations. It emphasizes on-device privacy and offline tool-calling capabilities.