LFM3-Omni-2B ported to run fully offline on iPhone at ~80 tok/s via custom WebGPU kernels
victormustar · x · 2026-11-05
The xenovacom team ported LiquidAI's LFM3-Omni-2B to run fully offline: first achieving the first true omni model running end-to-end locally in a browser — no separate ASR/TTS needed — with custom WebGPU kernels reaching 280 tok/s on an M4 Max, 90ms to first token and 140ms to first audio, accepting visual, audio, or text input and producing interleaved text and audio tokens.
They then brought it to an iPhone 18 Pro Max in a React Native app at 80 tok/s fully offline, and are asking whether to release it on the App Store. The team argues real-time local conversational AI is how intelligence gets democratized, and that on-device multimodal agents are about to get wild.
Related event: LFM3-Omni-2B Runs Fully Offline on iPhone at 80 tok/s(2 posts)→
More from Infra
- US AI scale faces two physical gates: delivered power and manufacturing — ingliguori · 2026-11-05
- Liquid AI ships LFM 3 models (1.7B/2B) running locally on iPhone Duo via MLX — helloiamleonie · 2026-11-05
- Yuma backs Bittensor subnet Fugal claiming GPT-5.5-level results at 1/6 the cost — markjeffrey · 2026-11-05
- Speculative decoding makes Qwen3.6/3.8 much faster, latest draft model adds ~10% — teortaxesTex · 2026-11-05
- webAI open-sources 8.8B TwIL-8B hitting 93% of frontier coding performance — dr_cintas · 2026-11-05
- Together Compiler optimizes new open models for production in hours, not weeks — togethercompute · 2026-11-05