← Retour au blog
tech 19 August 2026

Ornith-1.5: From Self-Scaffolding to Self-Improvement

Ornith-1.5 revolutionizes foundational model development with a comprehensive self-improvement loop, surpassing current performance standards.

Article inspired by the original source
Ornith-1.5: From Self-Scaffolding to Self-Improvement ↗ ornith.ai

Introduction: A Revolution in Progress

In August 2026, the AI world welcomed a major breakthrough with the launch of Ornith-1.5. This model does not just enhance existing capabilities; it redefines the standards of foundational model development through a comprehensive self-improvement loop. In this article, we explore how Ornith-1.5, building upon the success of Ornith-1.0, advances the field of AI towards uncharted territories.

From Self-Scaffolding to Self-Improvement

Ornith-1.0 introduced the concept of self-scaffolding, a method where the model structures itself to better adapt to new tasks. Ornith-1.5 goes further by incorporating a self-improvement cycle. The model not only proposes new tasks; it also generates task-specific scaffolds and produces solutions for reinforcement learning. This process continuously creates new learning experiences from which the model can improve.

Impressive Performance

Ornith-1.5 is available in three model sizes: 397B MoE, 35B MoE, and 9B dense. Each of these versions is designed for robust general-purpose intelligence across reasoning, agentic, and coding tasks. The 397B model scores 86.1 on Terminal-Bench 2.1 and 56.0 on DeepSWE, competing with Claude Opus 4.8 while outperforming open-source models of similar size such as GLM-5.2 and DeepSeek-V4-Flash-0731.

Technological Innovation

Ornith-1.5's learning cycle unfolds in three stages: task generation, scaffold construction for these tasks, and solution rollouts. Unlike traditional models that rely on predefined tasks and manual harnesses, Ornith-1.5 continually generates new training tasks, discovers effective strategies for solving them, and improves its policy through reinforcement learning.

Implications and Future

The impact of Ornith-1.5 goes beyond performance improvements; it opens up unprecedented prospects for automation and continuous improvement in AI models. By enabling self-generation of tasks and self-optimization, Ornith-1.5 reduces dependency on predefined human tasks. This hints at a future where AI models could evolve autonomously, adapting to increasingly complex environments.

Conclusion

Ornith-1.5 marks a crucial step in the evolution of AI models. By combining self-scaffolding and self-improvement, it redefines what we can expect from AI. For decision-makers, entrepreneurs, and developers, understanding and leveraging these advancements is essential to remain competitive.

Let's discuss your project in 15 minutes.

Ornith-1.5 self-improvement AI models automation self-scaffolding
Deepthix newsletter · 100% AI · every Monday 8am

An AI agent reads tech for you.

Our AI agent scans ~200 sources per week and ships the best articles to your inbox Monday 8am. Free. One click to unsubscribe.

Visit the newsletter page →

Want to automate your operations?

Let's talk about your project in 15 minutes.

Book a call