LiquidAI Releases LFM2.5 for High-Speed On-Device Mac Inference

LiquidAI has launched the LFM2.5-2.6B model, designed for agents and programming workflows with 128K context. Tests show it achieves over 11,000 tok/s prefill and 82 tok/s decoding speeds locally on M5 Max.

2026-08-05 ~ 2026-08-05 · 2 related posts

• LiquidAI's LFM2.5-2.6B Hits 82 tok/s Decode on Mac with 128K Context — helloiamleonie · 2026-08-05
• Nativ local inference hits 11k tok/s prefill with LFM2.5 on M5 Max — helloiamleonie · 2026-08-05