MiniMax open-sources H3 video model, community runs it on Macs
MiniMax released open weights for its H3 video model, which generates 15s 768p clips in 13s and runs 14x faster on a single GPU. Within 48 hours, the community ran H3 on untested hardware like $280 gaming GPUs and fully-offline MacBooks, and the creator of Redis built a Mac inference engine.
How this story unfolded
4 weeks · 1 report · 9 community posts · from Jul 31
- Jul 31
- Aug 3
- Aug 4
- Aug 5
- Aug 10
- Aug 28
MiniMax by email
Get an email when MiniMax has news
No news that day, no email.
More stories today
- DeepMind panel discusses generative media SOTA and human eval
- Reddit users share useful MCP servers for Claude
- Why embodied AI hits an edge AI wall requiring new math
- HTMX CEO mandates 'No AI Fridays' to counter LLM cognitive debt
- OpenAI product lead Tara Seshan discusses persistent AI coworkers