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

- Ornith-1.0 is an open-source AI model family designed for "agentic coding" — autonomous use of tools like code editors, terminals, and web browsers to solve software tasks. - It was built via a self-improvement loop: the team used stronger models (GPT-4o, Claude Sonnet 4) to generate synthetic training data (code, terminal interactions, web tasks), then trained smaller models on that data. - The repo claims Ornith-1.0 matches or beats GPT-4o and Claude 3.5 Sonnet on the SWE-bench coding benchmark, while being much cheaper to run. - This matters because "agentic" AI (models that can act, not just chat) is a major frontier in AI research, and most leading agentic models are proprietary (OpenAI, Anthropic). Ornith is an attempt to create a competitive open alternative via synthetic-data distillation.