Ornith-1.0

### TL;DR

Ornith-1.0 is a family of self-improving, open-source large language models (LLMs) designed for agentic coding tasks. Ranging from compact 9B Dense models suitable for edge devices to 397B MoE models optimized for maximum performance, Ornith-1.0 achieves state-of-the-art results on various coding benchmarks. Built upon pretrained Gemma 4 and Qwen 3.5, it introduces a self-improving training framework that enables the model to generate both solution rollouts and the task-specific scaffolds guiding those rollouts, leading to higher-quality solutions without relying on human-designed harnesses. This innovative approach allows Ornith-1.0 to outperform leading open-source models of similar size, demonstrating strong agentic coding capabilities even in resource-efficient deployments.

Key Insights & Metrics

Key Features

• Self-improving training framework
• Open-source models ranging from 9B Dense to 397B MoE
• State-of-the-art performance on coding benchmarks
• Built upon pretrained Gemma 4 and Qwen 3.5
• Strong agentic coding capabilities in resource-efficient deployments

→ Related Releases

Discussion