DeepReinforce has released Ornith-1.0, their new self-improving family of open-source models designed for agentic coding. > Ornith-1.0 learns to write its own task scaffolds during training rather than relying on human-designed harnesses. > The 397B MoE flagship can match