Mystery Model Scores Over 80% on DeepSWE, Beating Fable and GPT-5.6-sol

kimmonismus · x · 2026-08-21

A mystery model achieved over 80% on 10 DeepSWE tasks, significantly outperforming Fable (65%) and GPT-5.6-sol (52%). The author speculates it might be from a Chinese company, possibly a new GLM or Kimi model.

Original post →

More from Models

Models channel →