Ornith-1.5 represents a major advance toward end-to-end self-improving foundation models. Building on the self-scaffolding framework of Ornith-1.0, Ornith-1.5 establishes a complete self-improvement loop where the model proposes new tasks, generates task-specific scaffolds, and produces solution rollouts for reinforcement learning. This continuous cycle creates new learning experiences that drive sustained capability gains.

The system spans three model scales: 397B MoE, 35B MoE, and 9B Dense. The flagship Ornith-1.5-397B achieves state-of-the-art performance among open-source models of comparable size, scoring 86.1 on Terminal-Bench 2.1 and 56.0 on Deep SWE. These results match Claude Opus 4.8 (85.0 and 59.0) while outperforming leading open-source models like GLM-5.2 and Deep Seek-V4-Flash-0731.

At the smaller end, Ornith-1.5-9B delivers remarkably strong results for its size, achieving 47.0 on Terminal-Bench 2.1 and 70.6 on SWE-Bench Verified. Its quantized Ornith-1.5-9B-Mobile version can be deployed directly on iPhone and Android devices while substantially outperforming larger models such as Gemma 4-31B and Qwen 3.6-35B.

Rather than relying on static training distributions or hand-engineered agent design, Ornith-1.5 continually expands its own curriculum and adapts problem-solving strategies through a closed self-improvement loop involving task generation, scaffold construction, and solution rollouts.