Yesterday, I looked at the technical report (published last November) for Motif 2 (12.7b), the predecessor model of Motif 3, which achieved comparable results to DeepSeek Pro. It's amazing. It surpasses even the Qwen3 32b model with more than twice the parameters. Yesterday's achievement doesn't seem to have come out of nowhere.
▶ Original source: https://arxiv.org/pdf/2511.07464
