Codex has a big problem with accumulated context on long running agentic tasks. It’s fixable but spelunking to diagnose and fix shouldn’t be a necessary concern for users.
GPT-6.1 Sol is officially unusable in Codex. It’s gotten so bad that after burning through roughly 50% of my weekly usage limit, I ended up switching back to Opus 5.5. Honestly, I’m really sad about this because I love OpenAI, but they’re doing everything so wrong…