AIHacker Newsabout 3 hours ago
Ornith-1.5: From Self-Scaffolding to Self-Improvement
1 min read
Ornith-1.5 introduces a self-improvement loop where models propose tasks, generate scaffolds, and produce rollouts for reinforcement learning. It spans 397B MoE, 35B MoE, and 9B dense scales. The 397B scores 86.1 on Terminal-Bench 2.1, matching Claude Opus 4.8. The 9B runs on mobile devices.
Level
Hype check
Tap to vote and see what everyone thinks.
Summary by ByteBrief
