GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2
Read in full here:
1 Like
Hmm interesting article and wonderful example/reasoning provided by the author.
Also today, I discovered GLM 5.2 from Z. AI. This model competition is getting tighter by the day!
I think we’re reaching a point where architecture, training quality, inference efficiency, and post-training matter more than simply adding parameters. Larger models still tend to perform better overall, but the performance gains per additional billion parameters seem much smaller than they were a few years ago. It’ll be interesting to see whether the next major breakthroughs come from smarter model design rather than just making models bigger.