Ornith introduces 397B, 35B and 9B self-improvement models

Ornith introduced Ornith-1.5 in 397B MoE, 35B MoE and 9B dense versions. The company said its training system generates tasks, task-specific scaffolds and solution rollouts, then uses reinforcement learning across those stages. It reported that the 397B model scored 86.1 on Terminal-Bench 2.1 and 56.0 on DeepSWE, while the 9B version can be deployed on iPhone and Android devices. The company said results were averaged across five independent runs.