@MiaAI_lab Posts Her 100-Page Test of Ornith-1.5-122B
Very few broken files and some designs other models did not make; a TensorFold recipe drops tomorrow.
The promised test is in.
This morning @MiaAI_lab said she was blown away by Ornith-1.5-122B and would post her 100-HTML experiment soon. Before noon PT, she did. The rules were simple: look stunning, zero repetitive designs, full creative mode.
“This is no Opus 5.5, but for a model of THIS size, I’m impressed,” she writes. She says some recent frontier models did not wow her this much.
| Check | Result |
|---|---|
| Broken files | Very few |
| Repetitive files | Some, nothing major |
| Creativity | Some unique files other models did not produce |
| Suggested hardware | One DGX Spark or RTX 6000 PRO, with room to spare |
She published the results as-is, with no edits, alongside her earlier 100-HTML experiments with other models.
Smaller, with a recipe tomorrow.
Asked whether Ornith beats Qwen, she is measured. “Qwen3.8 next flash still has the edge, but this one is good too, and much smaller footprint with exl3.”
Her Ornith-1.5-122B recipe drops tomorrow, she says, and will be based on TensorFold.
Sources
- @MiaAI_lab on X · Oct 20, 11:23 AM PT
- @MiaAI_lab on X · Oct 20, 12:16 PM PT
- @MiaAI_lab on X · Oct 20, 3:40 PM PT