Hacker News Reader

Is there any technical analysis of why contex grows slower in Pi compared to codex and CC?

different system prompts, codex will have system prompt telling the model to gather a lot more context before starting work