Ornith-1.5-122B mixture-of-experts model trends on Hugging Face

The card describes a 122B MoE for single-GPU use and a quantized mobile variant, with vendor benchmark scores.

ornith-ai posted Ornith-1.5-122B, described as the most capable member of the Ornith-1.5 family and trending on Hugging Face. It extends Ornith-1.0, built on Qwen3.5 and Gemma4, by jointly optimizing task generation, scaffolds, and solution rollouts through reinforcement learning. The card calls the 122B mixture-of-experts model suitable for single-GPU deployment and says a quantized Ornith-1.5-122B-Mobile variant is edge-deployable. Published scores include 81.200 on TerminalBench 2.1, 80.000 on CyberGym, and 76.900 on SWE-bench Pro (no-hack).

Unconfirmed: Benchmark figures are the publisher's own model-card results, not an independent evaluation.