As Simon Willison notes, the Ornith-1.0 release exemplifies the current open-source playbook: remix capable base models under permissive licenses and claim state-of-the-art status on narrow benchmarks. The proliferation of variants—from 9B to a massive 397B MoE—suggests a scattershot strategy rather than a refined product. For developers, the value proposition appears to hinge on whether the promised 'agentic' capabilities materialize consistently in complex, real-world coding environments, not just in curated terminal sessions. The model's long-term viability may depend less on its benchmark scores and more on the stability and support of its obscure developer, DeepReinforce.