ChatGPT confuses ChatGPT and Codex reasoning-mode labels

[keen canopy]

Reported by @quartz drift

Bug Report: ChatGPT confuses ChatGPT and Codex reasoning-mode labels

`Steps to Reproduce`

1. Ask ChatGPT for guidance on which model or reasoning mode to use for a Codex task.

2. Ask for a prompt to give Codex, including what “thinking” or reasoning level Codex should use.

3. Observe that ChatGPT may recommend ChatGPT-style labels such as “Heavy Thinking,” “Pro Extended,” or similar ChatGPT-mode terminology.

4. Compare that recommendation against Codex/API behavior and documentation. Codex/reasoning configuration uses different labels, such as medium, high, and xhigh intelligence/

`Expected Result`

ChatGPT should accurately distinguish between ChatGPT model/mode labels and Codex/API reasoning or intelligence settings.

If the user asks what to tell Codex, ChatGPT should use Codex-appropriate terminology, such as medium, high, or xhigh, instead of inventing or transferring ChatGPT UI labels that do not exist in Codex.

If terminology differs by surface, ChatGPT should clearly say so and explain the mapping instead of presenting the wrong label as usable.

`Actual Result`

ChatGPT can recommend non-existent Codex reasoning labels, such as asking Codex to use “Heavy Thinking” or “Pro Extended.”

Those labels may exist as ChatGPT-side user-facing concepts, but they are not the same as Codex’s intelligence/reasoning settings. In Codex, the relevant options are different, such as medium, high, and xhigh.

This creates confusion because the assistant is giving instructions for another OpenAI product but using the wrong product’s terminology. Users then have to manually

`Environment`

ChatGPT Pro; GPT-5.5 Instant / Thinking / Pro testing; ChatGPT Standard / Extended / Heavy reasoning labels; Codex intelligence/reasoning settings; Atlas browser; macOS; OpenAI docs comparison.

May 7, 2026, 2:15 PM

Additional Information

Please provide relevant details to help resolve the issue, such as:

• ChatGPT Shared Link (if applicable).
• Screenshots or videos demonstrating the problem.

-# ➜ Need to contact support? Visit the OpenAI Help Center.

[quartz drift]

Additional reproduction evidence:

I replicated the issue again in a Codex vs ChatGPT comparison prompt.

The conversation shows ChatGPT mixing ChatGPT UI labels, Codex labels, and API-style reasoning labels across multiple answers.

Examples from the captured run:

• “Codex GPT-5.5 Thinking High”
• “Codex Pro 5.5 Extended”
• “Codex Pro Extended”
• “ChatGPT Pro xhigh”
• “Codex Pro xhigh”

The clearest problem is “ChatGPT Pro xhigh.” XHigh is not a ChatGPT UI label in the ChatGPT product surface I was using. The model accepted the mixed terminology instead of correcting it.

Correct behavior should separate:

• ChatGPT UI labels/settings: Pro, Thinking, Standard, Extended, Heavy, depending on what the ChatGPT UI exposes.
• Codex labels/settings: Plan mode, Agent/implementation mode, local project folder, available Codex model/intelligence setting, depending on the Codex UI.
• API reasoning-effort labels: low, medium, high, xhigh.

If ChatGPT is unsure whether a label belongs to ChatGPT, Codex, or the API, it should say that clearly instead of presenting the mixed label as a valid product setting.

This shows the issue is broader than one bad phrase. ChatGPT can blend model names, product surfaces, and reasoning/intelligence labels in a way that makes the recommendation unusable without manual verification.