ParentFull threaddarknoon·afaict OpenAI's instance is massively overloaded, you can see with the 32k context model actually being faster in practice rather than slowerView on HN