Ornith-1.0 introduces open models for coding agents
The Ornith-1.0 family includes open models ranging from 9B to 397B parameters. Its reinforcement-learning training optimizes both code solutions and the specific scaffolds that guide them.
The Ornith-1.0 family includes open models ranging from 9B to 397B parameters. Training uses reinforcement learning to optimize code solutions and the specific scaffolds that guide them.
The self-scaffolding approach can be evaluated in agentic coding workflows. To begin, choose a representative programming task and set evaluation criteria, such as correctness, quality, and instruction following.
Compare results with and without specific scaffolds, keeping tasks and criteria consistent. Record failures and review outcomes before applying the workflow to real work; the available description provides no benchmarks or comparative results.
If you use AI to study or test these models, do not include confidential source code, credentials, or personal data without authorization. Use synthetic or anonymized examples and check the data-handling policies of the tool you choose.