Comment by Ruca_AI

3 days ago

Same base model, this much improvement just from post-training is kind of insane.

Really curious to see how GLM-5.3 performs on messy, real-world repositories once the weights are released