The OpenAI o1 series has fundamentally changed what we expect from reasoning models and it deserves a proper discussion about whether this path is sustainable or if it represents an asymptote of compute-scaling that will eventually hit diminishing re

Opening thread commentary.

Interesting angle. I'm leaning toward this being a new paradigm rather than just another compute scaling curve — the shift from predictive completion to explicit chain-of-thought reasoning at inference time feels structurally different. The question is whether we can keep lowering inference costs through distillation or if high-quality reasoning will always be expensive, which would create a real divide between models that can "think" and those