invalid_request_error: reasoning item provided without required following item (Codex CLI 0.92)
Resolved 💬 5 comments Opened Jan 29, 2026 by ibabencu Closed Mar 5, 2026
💡 Likely answer: A maintainer (sayan-oai, contributor)
responded on this thread — see the highlighted reply below.
Summary
When using Codex CLI (v0.92) with GPT-5 Codex high reasoning, the API returns an invalid_request_error indicating a reasoning item was provided without a required following item.
Error
{
"error": {
"message": "Item 'rs_02df0e1c14fc6c3000697b676491008195a9819c02bdf5feab' of type 'reasoning' was provided without its required following item.",
"type": "invalid_request_error",
"param": "input",
"code": null
}
}
Environment
- Codex CLI: 0.92
- Model: GPT-5 Codex (high reasoning)
- OS: Windows
- Date/Time: 2026-01-29 (time + timezone not captured)
- Request ID: N/A
Steps to Reproduce
- Run Codex CLI with GPT-5 Codex high reasoning.
- Submit a prompt that triggers reasoning items in the request.
- API fails with the error above.
Expected
Request succeeds or error message explains the required following item format.
Actual
Request fails with invalid_request_error.
Additional Context
If you need the raw request payload, let me know what to capture from Codex CLI logs and I can provide it.
5 Comments
Hey @ibabencu, do you mind running
/feedbackand sharing the thread-id?i'll try finding the convo. i hope i can. yolo.
Edit: i'll ask codex to find it i have a SS
Hi! Adding redacted technical details (no secrets):
Thread ID: 019c092e-93f4-7e83-8c0b-9c60dabe27a6
Date/Time (UTC): 2026-01-29 ~14:00Z (from local Codex logs)
Codex CLI version: 0.92.0
OS: Windows
Provider: Azure OpenAI via Codex CLI
Model (deployment name): gpt-5.2-codex
Reasoning effort: high (via CLI profile)
Error (verbatim):
Repro (minimal):
1) Run Codex CLI with profile
high(reasoning effort = high).2) Send any prompt (I saw it on a normal user prompt; not tool-specific).
3) Response fails immediately with the error above.
Redacted request outline (best-effort; NOT a raw wire dump):
Not available in CLI logs by default: request_id, response headers, and raw request/response JSON. If you want me to re-run with a specific debug/logging flag to capture a fully redacted raw dump, tell me which flags/fields you need and I’ll provide it.
Redactions: No API keys, tokens, full base URL, or user paths included.
Additional detail: I ran Codex CLI with the YOLO option (flag: codex --dangerously-bypass-approvals-and-sandbox).
Similar to this https://github.com/openai/codex/issues/13480
Please report if you notice again