{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreic2zuqypzgbiqldot5o34gz55f5m6ewb4zczgmo3c6xgcd4nrytje",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mngfltcyjfu2"
  },
  "path": "/t/gpt-5-2-and-gpt-5-3-codex-have-been-sunset-in-codex-with-chatgpt-subscriptions/1382273?page=5#post_85",
  "publishedAt": "2026-06-03T23:44:17.000Z",
  "site": "https://community.openai.com",
  "textContent": "I need to test this further, but as far as I can remember Codex installed with Personality set in Pragmatic mode. I am using the default settings and don’t have any Custom instructions set either. I noticed yesterday that somehow my Personality setting had changed to Friendly. I moved this back to Pragmatic and 5.4 and 5.5 Low seem to be more compliant. The biggest difference that I notice between 5.3 Codex and 5.4/5 is that the newer models seem to make assumptions and kinda take the prompt with a grain of salt, like it knows better, It then reasons itself into taking an action that is counter to the original prompt. For example, using 5.3 I have been able to take screenshots of working pages, enabling it to notice issues and correct them itself. However 5.4/5 seems to have required major prompting to get its environment to a point where it can take screenshots reliably. One issue came up and I explained that a large amount of white space was in the screens, please check the screenshot and the model immediately assumed the code was wrong and started editing the code to “fix” the space. I have never seen 5.3 make this type of blunder, not even once.",
  "title": "GPT-5.2 and GPT-5.3-Codex have been sunset in Codex with ChatGPT subscriptions"
}