Back to feed
models20h ago
GPT Codex users: at what token count does quality actually degrade for you?
GPT Codex users report quality degradation around 180k tokens, with sloppy reasoning and dropped context. Users cap sessions at 170k tokens to maintain quality. You can expect performance to decline after 180k tokens, impacting real coding work. Quality issues arise from sloppy reasoning and wrong assumptions.
Key takeaways
- Quality degrades around 180k tokens.
- Users cap sessions at 170k tokens.
- Sloppy reasoning and dropped context are common issues.
GPT Codex users report quality degradation around 180k tokens, with sloppy reasoning and dropped context. Users cap sessions at 170k tokens to maintain quality. You can expect performance to decline after 180k tokens, impacting real coding work. Quality issues arise from sloppy reasoning and wrong assumptions.
Key takeaways
- Quality degrades around 180k tokens.
- Users cap sessions at 170k tokens.
- Sloppy reasoning and dropped context are common issues.