Ornith-1.0: self-improving open-source models for agentic coding

Ornith releases open-source self-improving models for agentic coding tasks, with Ornith-1.5 representing the latest generation that implements an end-to-end self-improvement loop where the model proposes tasks, generates scaffolds, and produces solution rollouts for reinforcement learning. Available in multiple sizes (9B, 35B, and 397B parameters), both Ornith-1.5 and Ornith-1.0 are MIT-licensed and demonstrate competitive performance on coding benchmarks including SWE-bench, Terminal-Bench, and various agentic evaluation tasks. The models are globally accessible and free from regional restrictions, with detailed benchmarks and weights available on HuggingFace.