Wtf! Ben ran this mystery model through 10 DeepSWE tasks and it scored over 80%, versus 65% for Fable and 52% for GPT-5.6-sol! this is insane. Probably a chinese company. Either a new GLM or Kimi model, I reckon.