The LFM3-Omni-2B model can now run entirely within a web browser using WebGPU, eliminating the need for separate audio and visual processing models. This advancement allows for local, end-to-end processing of visual, audio, or text inputs, producing interleaved text and audio tokens. The model achieves approximately 280 tokens per second on an M4 Max chip, with a first token latency of 90 milliseconds and first audio latency of 140 milliseconds. AI
IMPACT Enables more accessible, local execution of multimodal AI models directly within web applications.
RANK_REASON This is a technical demonstration of a model running in a new environment, not a core AI release from a major lab.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
