AI BriefingOrnith-1.5-122B Introduced with End-to-End Self-Improvement Loop
·2026.10.20 23:19
Key point
The model jointly optimizes task generation, scaffold construction, and solution rollouts via reinforcement learning to replace fixed human-curated training sets.
Details
- Ornith-1.5-122B is a major step toward foundation models built through end-to-end self-improvement, extending the previous Ornith-1.0 architecture.
- The new model expands the self-improvement loop to jointly optimize task generation, scaffold construction, and solution rollouts, moving beyond fixed human-curated tasks and manually designed harnesses.
- The system continuously generates new training tasks, discovers effective solving strategies, and improves the policy through reinforcement learning.
- Ornith-1.0 was developed on top of Qwen3.5 and Gemma4 with additional continued pretraining, mid-training, and post-training.
- GGUF versions are available on Hugging Face under ornith-ai and unsloth accounts.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.