Edge0-145B: Frontier AI Capabilities on Your Own Computer
Web page · https://edge0.ai/blog/introducing-edge0-145b-frontier-ai-capabilities-on-your-own-computer
First seen by Oriel on .
AI summary
The page announces open-source Edge0-145B and Edge0-120B, claiming frontier-like coding, reasoning, and agent capabilities locally through a unified inference and deployment stack.
- Edge0-145B reportedly uses 29GB peak memory, exceeds 80 tokens per second during decoding, processes prompts above 1,100 tokens per second, and supports 128K–256K context.
- Its architecture shares matrices across sparse experts, while pipelined GPU, system memory, and SSD access and CPU-GPU scheduling reduce memory movement; approximately 10B parameters compute per step.
- Across six benchmarks, Edge0-145B averages 72.8, compared with 71.7 for Qwen3.8-Flash-Next, whose estimated standard 4-bit footprint is approximately 70GB.
- The model can plan, write, execute, debug, and visually inspect local workflows involving video, Blender scenes, games, drone control, and PCB design.
- Reported comparisons and memory figures combine internal evaluations, measured Edge0 results, and estimated baseline footprints, with complete reproduction details promised in a technical report.
#on-device ai #inference #ai agents #sparse experts
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Hacker News: 345, latest observation ; +338 over observed period since
Comments
Comment count reported by the source.
- Hacker News: 105, latest observation ; +105 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hacker News | 50475713 | https://edge0.ai/blog/introducing-edge0-145b-frontier-ai-capabilities-on-your-own-computer Read discussion |