ornith-ai/Ornith-1.5-35B-A3B model card

Fetched 2026-10-18. Weights repo created 2026-08-18 (HF API createdAt 2026-08-18T06:24:38Z). Card nameplate: mid-size MoE of the Ornith-1.5 family, ~35B total, ~3B activated per token, 262,144 context, MIT. The card's own benchmark table is a vendor claim, averaged over five runs: SWE-bench Verified 74.2, GPQA Diamond 88.6, Terminal-Bench 2.1 (Terminus-2) 56.4, AIME 2026 96.7. HF safetensors metadata reports 35,951,822,704 BF16 parameters and total_size 71,903,645,408 bytes. The card says the model continues from Ornith-1.0, which itself was developed on top of Qwen3.5 and Gemma 4; the config model_type is qwen3_5_moe.