I think they distilled from GPT after Claude tightened things up. It used to feel closer to Claude in personality, now it's stuck somewhere between Claude and GPT. Looks like their RL also leaned hard into viral stuff like 3D and neglected the rest, and the compute shortage might be limiting what they serve, like cutting the thinking budget. I'll wait for the open weights and test it in Devin before making a final call.

The free SWE-2 offer inside Devin just got extended to December 1. You literally have a SOTA model you can use with zero limits and ship at a level nothing else comes close to. Huge win for everyone using Devin in November.