Comment by manmal

The benchmarks don’t look _that_ much better than o3. Does that mean Pro models are just incrementally better than base models, or are we approaching the higher end of a sigmoid function, with performance gains leveling off?