GPT-5 Bugs? A Complete Guide to Reporting Issues to OpenAI
Z
Zack Saadioui
8/13/2025
So, the new GPT-5 model is out, & it’s been a whirlwind. Honestly, it feels like the entire AI community has been waiting for this one. OpenAI promised a lot – a huge leap in reasoning, coding capabilities that could write entire apps, & a more unified, intuitive user experience. & in many ways, it delivered. The benchmarks are impressive, with GPT-5 acing things like the AIME 2025 math competition & showing much-improved performance in complex health-related conversations.
But here's the thing about a massive, cutting-edge AI release: the real test begins when it's in the hands of millions of users. & as expected, the initial rollout has been a bit of a bumpy ride. If you've been playing around with GPT-5, you might have noticed some… quirks. Maybe it's struggling with tasks that GPT-4 handled with ease, or you've run into weird formatting issues. You're not alone. The internet is already buzzing with discussions about GPT-5's little (and sometimes not-so-little) problems.
The good news is, this is a normal part of the process. Every major software release has bugs, & with something as complex as a new foundational AI model, it's pretty much guaranteed. The even better news is that you can play a crucial role in making GPT-5 better. By reporting the bugs & issues you encounter, you’re not just helping yourself; you’re helping the entire community & giving OpenAI the data they need to iron out the kinks.
So, how do you do it? Where do you even start? Don't worry, I've got you covered. Here’s a detailed guide on how to properly report bugs & issues with GPT-5.
The Unofficial Town Square: Community Bug Reporting
Before we get into the official channels, let's talk about the power of the community. Right now, a lot of the conversation around GPT-5 bugs is happening on platforms like Reddit, X (formerly Twitter), & Discord. These are fantastic places to see if others are experiencing the same issues as you.
There are already dedicated threads & even a community-run GPT-5 Bug & Issue Tracker. These unofficial hubs are great for a few reasons:
Validation: You can quickly see if your issue is a one-off or a widespread problem.
Workarounds: Sometimes, other users have already figured out a temporary fix or a way to avoid the bug.
Visibility: Popular threads & community trackers can sometimes get the attention of the folks at OpenAI faster than individual reports.
One of the cool things about these community trackers is that they often pre-format tweets or posts for you, making it super easy to share the bug with relevant hashtags like #gpt5bug. This helps to centralize the discussion & create a louder signal.
However, a word of caution: while these community platforms are great for discussion, they are not a substitute for official bug reports. A moderator on the OpenAI forums has pointed out that making comments in the forums won't necessarily get a bug fixed. So, while it's a good first step, don't stop there.
The Official Channels: Getting Your Report to OpenAI
When you're ready to submit an official report, you have a couple of options. It can be a little hidden, but here's how to find the right place.
The OpenAI Help Center
This is the primary, official way to report a bug. It's a bit of a process, but it ensures your report goes into their system. Here's a step-by-step guide:
Head over to the OpenAI Help Center at
1
https://help.openai.com
.
Look for a chat-like icon, usually in the bottom right corner. Click on it.
You'll likely be greeted by a chatbot. You'll need to navigate through a few options to get to the right place. Look for options like "Ask a question" or something similar.
You'll be presented with a series of choices. Select the options that best describe your issue. You'll eventually get to a point where you can submit the details of the bug.
It's not the most straightforward process, & honestly, it could be a bit more user-friendly. But for now, this is the main pipeline for getting your bug report into the official queue.
The OpenAI Discord
OpenAI has an official Discord server, & on that server, there's a
1
#bug-reports
channel. This is another place where you can post your findings. While some users have noted that they haven't seen a ton of direct responses from OpenAI employees in this channel, it's still a good place to log the issue. The more places you can document the problem, the better.
Writing a Bug Report That Actually Gets Read
Now for the most important part: how to write a bug report that is clear, concise, & actionable. A well-written bug report can be the difference between a bug getting fixed quickly & it languishing in the backlog. Here are the key elements of a great bug report, based on best practices from the software testing world:
1. A Clear & Concise Title
Your title is the first thing a developer will see. It should summarize the bug in a single sentence. A good title is specific & easy to understand at a glance.
Bad Title: "GPT-5 is broken"
Good Title: "GPT-5 generates incomplete JSON output for nested API workflows"
2. A Detailed Description
This is where you elaborate on the title. Explain the problem in more detail. What were you trying to do? What happened instead? Why is this a problem?
Example: "I was using the GPT-5 API to generate a JSON object with multiple nested levels. I provided a clear prompt with the desired structure, but the model consistently cuts off the output before the JSON is complete. This is a problem because it breaks my application's parsing logic."
3. Steps to Reproduce
This is arguably the MOST critical part of your bug report. If a developer can't reproduce the bug, they can't fix it. Be as specific as possible. List the exact steps you took, in order.
Example:
Called the GPT-5 API with the
1
gpt-5-pro
model.
Used the following prompt:
1
[insert your exact prompt here]
Set the
1
temperature
to 0.2 &
1
max_tokens
to 2000.
Observed that the API response was a JSON string that was cut off mid-way through.
4. Expected vs. Actual Results
Clearly state what you expected to happen & what actually happened. This removes any ambiguity.
Expected Result: "I expected the API to return a complete, valid JSON object that matches the structure I requested in the prompt."
Actual Result: "The API returned an incomplete JSON string. The closing brackets for the main object were missing, & the last nested object was only partially generated."
5. Provide Visual Evidence
A picture is worth a thousand words, & in bug reporting, it can be worth a thousand lines of code. Include screenshots, screen recordings, or logs whenever possible. If you're reporting an API issue, include the full request & response, including headers. For UI bugs, a screenshot with the problematic area highlighted is incredibly helpful.
6. Environment Details
The environment you're using can have a big impact on how the model behaves. Include as much information as you can about your setup.
For UI bugs: What browser are you using (Chrome, Firefox, Safari)? What is your operating system (Windows, macOS, Linux)? Are you using the web interface or a mobile app?
For API bugs: What programming language & library are you using to make the API call?
7. Severity & Priority
This helps developers triage the bug. Severity refers to the impact of the bug, while priority is about how quickly it needs to be fixed.
Severity Levels (examples):
Blocker: Prevents you from using a key feature entirely.
Major: A significant issue that has a major impact on functionality.
Minor: A small issue with a limited impact.
Trivial: A cosmetic issue, like a typo in the UI.
Priority Levels (examples):
High: Needs to be fixed immediately.
Medium: Should be fixed in the next release.
Low: Can be fixed when time allows.
The Bug Bounty Program: For Serious Security Issues
It's important to distinguish between a regular bug & a security vulnerability. A regular bug might be something like GPT-5 giving a wrong answer or formatting code incorrectly. A security vulnerability is much more serious – it's a flaw that could be exploited to compromise the system, access private data, or cause harm.
OpenAI has a bug bounty program for precisely these kinds of issues. This program is run in partnership with Bugcrowd, a popular bug bounty platform. If you believe you've found a security vulnerability, you should report it through this program. The rewards can be substantial, ranging from a few hundred dollars for low-severity issues to as much as $20,000 or even $100,000 for exceptional discoveries.
However, OpenAI is clear that model safety issues – like the model generating biased or harmful content – are not typically a good fit for the bug bounty program. These issues are better reported through the regular channels, as they often require significant research to address.
A Note on Business Applications & Feedback Loops
For businesses that are integrating GPT-5 into their products & workflows, the feedback loop is even more critical. If you’re a company using GPT-5 to power a customer-facing tool, any bugs or performance issues can directly impact your customer experience. This is where having a robust system for collecting & analyzing user feedback becomes essential.
This is a great example of where a tool like Arsturn can be incredibly valuable. Imagine you've built a customer support chatbot using GPT-5. Your customers are interacting with it 24/7. Arsturn can help you create a custom AI chatbot that not only provides instant support but also acts as a feedback collection mechanism. You can train the chatbot on your own data to understand common customer issues & questions. When a user has a problem with the chatbot's response, you can build in a simple feedback mechanism, like a thumbs up/down button, that flags the conversation for review. This creates a continuous feedback loop, allowing you to identify & address issues with your GPT-5 implementation in near real-time.
For businesses focused on lead generation or website engagement, this is also key. Arsturn allows you to build no-code AI chatbots that can have personalized conversations with your website visitors. If a visitor is getting frustrated with the chatbot's responses, that's a potential lead lost. By analyzing these conversations, you can pinpoint where the GPT-5 powered chatbot is falling short & refine its prompts & behavior. It’s all about creating a more meaningful & effective connection with your audience.
Final Thoughts
The launch of a new GPT model is always an exciting time, but it's also a collaborative effort. The initial bugs & issues are not a sign of failure, but rather an opportunity for the community to come together & help shape the future of this technology. By taking the time to write clear, detailed bug reports, you're not just a user – you're a contributor.
So next time you run into a weird issue with GPT-5, don't just get frustrated. Document it, report it, & know that you're playing a vital role in the evolution of AI.
Hope this was helpful! Let me know what you think, & feel free to share your own experiences with GPT-5 in the comments.