Have any of you encountered a situation where GPT seems to be "dumbing down"? I'm currently using the Plus version, but even though it says "ChatGPT-5Thinking" model, the responses are actually from GPT-4. Additionally, I can't use "Deep Research" either. How can this be resolved?
#Have you encountered GPT "dumbing down" issues? How to solve it?
dull scroll
plush graniteBOT
Hi @dull scroll. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:
Yes, some users have reported situations where GPT models appear to "dumb down"—that is, responses seem less sophisticated or revert to older model behaviors. Here’s what the knowledge sources say about these issues and possible solutions:
1. Model Switching and Perceived "Dumbing Down"
-
Automatic Model Switching: In ChatGPT, especially with GPT-5, the system can automatically switch between different reasoning modes (e.g., GPT-5 Chat, GPT-5 Thinking, or even legacy models like GPT-4.1 mini) based on usage limits, demand, or prompt complexity. For example, if you hit your GPT-5 or GPT-5 Thinking message cap, ChatGPT will automatically switch you to a "mini" or legacy model, which may feel less capable or "dumber" than the full model. This is especially common on the Free and Plus tiers, which have stricter usage limits than Pro or Business plans. When this happens, you may notice a drop in response quality or a model name that doesn’t match your selection GPT-5 in ChatGPT.
-
Legacy Model Fallback: If you see responses that seem to come from GPT-4 or earlier, it’s likely due to hitting usage limits or the system auto-switching to a legacy model. This can also happen if you open an old conversation that was started with a legacy model; ChatGPT will try to map it to the closest GPT-5 equivalent, but outputs may differ GPT-5 in ChatGPT.
2. Deep Research and Feature Access
- Deep Research Availability: The "Deep Research" feature is only available on certain paid plans and may require enabling specific connectors or settings. If you’re on Plus and can’t access Deep Research, it may be due to plan limitations or feature rollout timing. Pro and Business plans have broader access to advanced features like Deep Research ChatGPT Rate Card.
3. Troubleshooting and Solutions
-
Check Usage Limits: If you’re on Plus, you have a cap of 160 GPT-5 messages per 3 hours and up to 3,000 GPT-5 Thinking messages per week. Exceeding these will cause automatic fallback to a less capable model, which can feel like "dumbing down." Wait for your limit to reset or consider upgrading to Pro or Business for higher or unlimited usage GPT-5 in ChatGPT.
-
Manual Model Selection: Use the model picker to manually select GPT-5 or GPT-5 Thinking. If you want to ensure you’re not using a fallback model, check your current model in the picker. If you don’t see the expected model, you may have hit a usage cap or need to enable "Show additional models" in settings GPT-5 in ChatGPT.
-
Reasoning Effort Settings: If responses seem shallow, try toggling the "thinking time" or reasoning level in the message composer (available for Plus, Business, and Pro users). Choosing "Extended" or "Heavy" can prompt deeper, more thoughtful answers ChatGPT Release Notes.
- Troubleshooting Overthinking/Underthinking: If the model is underperforming (either overthinking or underthinking), you can adjust your prompts or use the reasoning effort parameter to encourage more or less depth in responses. See the GPT-5 Troubleshooting Guide for best practices.
4. Why Model Names May Not Match
- Model Self-Identification: Models do not inherently know their own names. If you ask "What model are you?" the answer may be inaccurate unless specified in the system prompt. This is a known behavior and not a sign of malfunction OpenAI Community Forum.
Summary:
If you’re experiencing "dumbing down," it’s likely due to usage limits causing fallback to a less capable model, or the system auto-switching based on your plan. To resolve this, monitor your usage, manually select your desired model, and consider upgrading your plan if you need more consistent access to advanced features like Deep Research. Adjusting reasoning settings can also help improve response quality. If issues persist, contact OpenAI support for further assistance.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
dull scroll
The deep research limit is still there, and I’m using the paid version. It seems like the issue hasn’t been resolved. Have any of you encountered this situation?