DeepReinforce.AI Ornith 1.0

Ornith 1.0 397B is a 397B parameter sparse Mixture-of-Experts model designed for high-performance agentic coding and software-engineering workflows. Built on Qwen3.5, it uses a 60-layer hybrid architecture combining linear and full attention, with a 4,096 hidden size, 32 attention heads, 2 KV heads, and 512 experts activating 10 per token. Approximately 17B parameters are active per token, with a 262K-token context window and multimodal support. Its self-improving RL framework jointly optimizes solution rollouts and scaffolds, enabling stronger search trajectories and higher-quality coding solutions.

Ornith 1.0 35B is a 35B parameter sparse Mixture-of-Experts model designed for agentic coding and software-engineering workflows. Built on Qwen3.5, it uses a 40-layer hybrid architecture combining linear and full attention, with a 2,048 hidden size, 16 attention heads, 2 KV heads, and 256 experts activating 8 per token. Supporting a 262K-token context window and multimodal inputs, it targets coding agents, repository-level tasks, and tool-driven workflows. Its self-improving RL framework jointly optimizes solution rollouts and their scaffolds for stronger search trajectories.

Ornith 1.0 9B is a 9B parameter dense multimodal model designed for agentic coding and software-engineering workflows. Built on Qwen3.5, it uses a 32-layer hybrid architecture combining linear and full attention, with a 4,096 hidden size, 16 attention heads, and 4 KV heads. Supporting a 262K-token context window, the model is optimized for coding agents, repository-level tasks, and tool-driven workflows. Its self-improving RL framework jointly optimizes solution rollouts and the scaffolds that guide them, enabling stronger search trajectories and higher-quality solutions.