← The Polylog AI Intelligence Brief
Morning Edition · Tuesday, August 11, 2026Published at 2:26 AM EDT · New York
A 33-Billion-Parameter Video and Audio Model Now Runs on a MacBook, Slowly
Salvatore Sanfilippo has written a native Metal inference engine for MiniMax-H3, while a separate MLX port needs about 115 gigabytes of weights and takes just under 45 minutes per clip on an M5 Max.
MiniMax-H3 is a 33-billion-parameter joint video and audio diffusion model, and two independent efforts have now brought it to Apple Silicon without using PyTorch. Salvatore Sanfilippo, the creator of Redis, is building h3.c, a native infer…
Continue the AI Intelligence Brief
Track frontier labs, chips, export controls, model releases, regulation, and AI infrastructure.
- 5 AI intelligence signals a day
- Frontier labs, compute, and chips
- Model releases and AI infrastructure
- Source-grounded analysis with confidence labels
The Global Intelligence Brief stays free.
Part of a tracked trend
AI Inference Shifts to Consumer Devices
Over the next 3-6 months, smaller efficient architectures and inference-cost optimizations push capable AI off the cloud and onto laptops, phones, and mobile NPUs.
Townsquare
Start a discussion in Townsquare.
More from this edition
- OpenAI Ships a Purpose-Trained Offensive Security Model to Vetted Partners Only
- Meta Superintelligence Labs Returns to Open Weights With a 30B Agent Model Under Apache 2.0
- Hassabis Leaves the DeepMind Chief Executive Role as Jeff Dean Departs Google
- Agent Tooling Fights Its Own Token Overhead as Qwen Ships a Harness-Agnostic Multimodal Layer
- Three New Papers Attack the Weak Points of Multimodal Models at Once
- Researchers Test Whether Claude Models Write Code That Passes a SOC 2 Audit
- Schmidt Puts Frontier Training Runs at $200m to $500m With a Six-Month Useful Life
Comments
0No comments yet.