What happened
DeepReinforce-AI released Ornith-1.0, a suite of self-improving open-source agentic coding models, available in 9B-Dense, 31B-Dense, 35B-MoE, and 397B-MoE variants. Post-trained on Gemma 4 and Qwen 3.5, Ornith-1.0 achieves state-of-the-art performance among open-source models on coding benchmarks including Terminal-Bench 2.1 and SWE-Bench. The models employ a self-improving training framework using reinforcement learning, are MIT licensed, and the 9B model fits on a single 80GB GPU, supporting a 256K context window.
Why it matters
Widespread access to self-improving agentic coding models will reduce development timelines for platform engineers and founders. The MIT licence and single-GPU deployment for the 9B model lower infrastructure costs, enabling broader adoption of advanced coding agents. This offers a direct mechanism to cut the high LLM coding costs recently reported, shifting the metric for internal development and allowing teams to run complex agentic workflows on more accessible hardware.
