While closed-source giants are still competing over clusters worth tens of billions of dollars, this week the open-source community delivered two major releases that emphatically shattered the “compute is everything” mindset:
1. MiniMax open-sourced MiniMax-H3-Image, a native 4K image-generation model that produces 16-megapixel images (4096×4096) in a single pass without upscaling, with consistent results across up to 16 reference images.
2. Edge0 released Edge0-145B, an open-source MoE coding model with 145 billion parameters, with its framework redesigned specifically for local inference on a single machine.
When top-tier visual assets and 145B coding capabilities can both run on a single GPU at the edge, the barriers to the rise of decentralized physical infrastructure networks (DePIN) and AI agents for super-individuals are being swept away overnight. #AI #开源模型 #DePIN
1. MiniMax open-sourced MiniMax-H3-Image, a native 4K image-generation model that produces 16-megapixel images (4096×4096) in a single pass without upscaling, with consistent results across up to 16 reference images.
2. Edge0 released Edge0-145B, an open-source MoE coding model with 145 billion parameters, with its framework redesigned specifically for local inference on a single machine.
When top-tier visual assets and 145B coding capabilities can both run on a single GPU at the edge, the barriers to the rise of decentralized physical infrastructure networks (DePIN) and AI agents for super-individuals are being swept away overnight. #AI #开源模型 #DePIN