{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreiaynkwwl3jie77i5ixyf7jc6fbhqfzecokycwqscp22unoioklcsa",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mmvm4b2t7kn2"
  },
  "path": "/t/stealth-model-swap-gpt-5-5-high-claims-knowledge-cutoff-is-june-2024/1381918#post_6",
  "publishedAt": "2026-05-28T07:42:04.000Z",
  "site": "https://community.openai.com",
  "textContent": "Fair.\n\nThen this may come down to two things:\n\n  * It’s somehow picking up a fragment of a system prompt (possibly from some layer of the architecture)\n  * Some model routing is going on? It’s entirely possible “GPT 5.5” is a collection of models and there is a router sitting in front of them that decides which actual LLM is despatched depending on task profile, availability, load etc. However this may be less likely if you are seeing more consistency not less.\n\n\n\nIt really would be nice for OpenAI to tell us what is happening under the hood …",
  "title": "Stealth Model Swap? GPT-5.5 High Claims Knowledge Cutoff is June 2024"
}