Liquid AI LFM 2.5-230M: 230M Model Beats 1B Transformer on Edge
Liquid AI LFM 2.5-230M outperforms models 4x its size on data extraction and runs at 42 tok/s on a Raspberry Pi. Here is ...
Baidu Unlimited-OCR: One-Shot PDF Parsing Is Here
Baidu's Unlimited-OCR replaces the page-by-page OCR loop with constant KV cache via R-SWA. MIT licensed, ...
MiniMax M3: Open-Weight Frontier Model at 5% of Opus Cost
MiniMax M3 is an open-weight model with a 1M-token context window, 59% SWE-Bench Pro, and ...
MLX + JACCL: Distributed AI Training Over Thunderbolt 5
Apple shipped JACCL at WWDC 2026 — an open-source collective communication library that enables distributed ...
ZAYA1-8B: Run a Frontier Reasoning Model Without NVIDIA
ZAYA1-8B is an Apache 2.0 sparse MoE reasoning model trained entirely on AMD Instinct MI300X ...
NVIDIA Nemotron 3 Nano Omni: Run It Locally Now
NVIDIA Nemotron 3 Nano Omni is a 30B open multimodal model (3B active) with video, audio, and image support. Here is how to ...
Apple Evaluations Framework: Measure iOS AI Feature Quality
Apple's new Evaluations framework in Xcode 27 gives iOS developers a first-party way to measure ...
MLX Distributed Training with JACCL: Multi-Mac LLM Clusters, Explained
Apple shipped JACCL with macOS 26.2 — a distributed backend for MLX that runs trillion-parameter ...
Apple Image Playground API: Add Photorealistic AI to Your iOS 27 App
Apple's Image Playground now generates photorealistic images in iOS 27. Learn the new imagePlaygroundSheet API, ...
MLXLanguageModel: Run Hugging Face Models via Foundation Models
Apple's MLXLanguageModel runs any Hugging Face mlx-community model through Foundation Models on macOS 27. Same ...