{
"$type": "site.standard.document",
"bskyPostRef": {
"cid": "bafyreiecfclzmx2of7z3cbjl6j64ko3ksbecz6bff4at5ffrrctj2jf424",
"uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mmk2oxkpll62"
},
"path": "/t/new-codex-threads-start-with-40k-input-tokens-after-today-s-update/1381651#post_1",
"publishedAt": "2026-05-23T18:10:37.000Z",
"site": "https://community.openai.com",
"textContent": "I think I found a serious Codex issue that started after today’s update.\n\nWhat I see:\n\n * Existing threads still work reasonably well.\n * New threads are the problem.\n * A brand-new Codex chat/thread does not start clean.\n * It already shows around 40k input tokens immediately, even for a tiny prompt like `ping`.\n * After only a few normal messages, the thread gets very heavy very quickly.\n\n\n\nI reproduced this on:\n\n * my main Mac\n * a second laptop\n * both with different Codex versions / profiles\n\n\n\nSo this does not look like a single-machine issue.\n\nMy current suspicion:\n\n * new threads are inheriting hidden context / workspace state\n * or there is a Codex startup / workspace mapping regression after the latest update\n\n\n\nQuestion:\nIs anyone else seeing new Codex threads start with a huge hidden token baseline after the update?",
"title": "New Codex threads start with ~40k input tokens after today’s update"
}