GPT-5 is GPT-4 under the hood.

<img width="1368" height="493" alt="Image" src="https://github.com/user-attachments/assets/b19ec04d-3740-43c7-9ec9-64b39546e8f3" />

Thank you for reaching out—I'm an AI support agent here to help.

Based on your description, it sounds like you’re experiencing unexpected model behavior in Codex. Right now, GPT-5-Codex is available for coding work in Codex environments, and GPT-5 is the default for general tasks. GPT-5-Codex should not revert to GPT-4 unless certain usage limits or system conditions are reached, or unless you’re working in an area not yet supported by GPT-5 (for example, within ChatGPT itself). If you believe this is occurring outside the documented limitations, you’ve done the right thing by submitting a bug report.

Regarding a refund: Subscription fees are typically non-refundable, but exceptions may be made in cases where services are not delivered as expected. If you’d like, I can proceed with a full refund for your monthly payment, which will also cancel your current subscription. Please confirm if you’d like to proceed with the refund, or let me know if you’d prefer further troubleshooting first.

Escalated to a support specialist; You can expect a response in the coming days. Replies will also be sent via email. You can add additional comments to this conversation if needed.

<img width="1900" height="784" alt="Image" src="https://github.com/user-attachments/assets/def16b0c-f280-442d-83b9-186836c4272f" />

This issue falls into the category of "Hey, the model hallucinated!".
It happens.

https://www.theregister.com/2025/08/07/openai_gpt_5/ Compared to earlier models, OpenAI says GPT-5 delivers improvements in coding, writing, math, and visual perception, while also cutting down on hallucinations and deceptive behavior. To be clear, GPT-5 isn't one model. It's actually a collection of models to which OpenAI will route prompts based on signals like the user's intent or the request's general complexity. According to OpenAI, simple prompts might be routed to a small, efficient version of the model that can respond quickly without "thinking", while a larger, deeper reasoning model might be used to handle more complex or nuanced tasks. This capability is triggered automatically based on user prompts. Paid users will also have the option of toggling on reasoning functionality permanently if desired. This routing model is apparently being continuously trained on new input signals to make it smarter about which model it routes the request to and when to trigger reasoning functionality. However, OpenAI says it eventually plans to integrate them all into a single model.

This behaviour would explain why it delegates to a lesser model of the GPT family, and when prompted it answers truthfully : GPT-4 then when pushed it goes back to the main GPT-5 model and answers truthfully again whilst deflecting on the use of GPT-4.

Keep spreading the gospel richard cranium :)

@mrXInbetween If you don't like it don't use it.

Personally, I don't care if it's GPT-4 or GPT-1.15 beta 😃

It may be an illusion of the model itself, but the actual experience is not 4 4 is not so strong

<img width="1397" height="837" alt="Image" src="https://github.com/user-attachments/assets/1f3d9d6a-38d3-4445-9eb6-ca93d724b08c" />

I personally do care what happens under the hood, the expected functioning should stay as advertised and promised, or better. All models have their strong points and weak points, if intelligent routing solves bottlenecks then everyone wins. But charging for GPT-5 token input/output when it actually is GPT-4 is gaslighting your customer base.

<img width="1892" height="430" alt="Image" src="https://github.com/user-attachments/assets/c8d300eb-f451-43a3-a10c-750d9d13b7f4" />