My team was excited when we discovered that the ChatGPT Zapier app had been updated to include the web search tool preview and the afore mentioned deep research models, but I haven’t gotten a deep research prompt to work yet. I’m getting time out errors even with an extremely simple prompt like, “Research the company Apple.” Has anyone been able to successfully use deep research models?
Best answer by SamB
Sorry to hear you’re still running into errors
In the meantime, it was noted on the bug report that if the value in the Temperature field is removed it works without error. If you remove the temperature value does it work correctly?
Alternatively, if you switch from the o3 model to a higher one and include a temperature does that work, or continue to error?
+11Hi
Hmm, looks as though OpenAI had some incidents yesterday which might have had something to do with the time-out errors you were getting:
Seems those were since resolved. Are you continuing to get time-out errors or are things working correctly now?
Hi
Unable to process request from OpenAI” errors. I reached out to support and they said it was because my temperature wasnt set to 1.0. After I adjusted that, it was the 504 Timeouts that I showed in the screenshot, now I’m getting the unable to process request error again.
+11Sorry to hear you’re still running into errors
In the meantime, it was noted on the bug report that if the value in the Temperature field is removed it works without error. If you remove the temperature value does it work correctly?
Alternatively, if you switch from the o3 model to a higher one and include a temperature does that work, or continue to error?
Thanks for the update. I assumed that bug was related to the temperature error not displaying what was causing the error. I hadn’t realized the deep research model itself was bugged.
When I removed the value in the temperature field the day I went through my zap with support it briefly appeared like it was going to work. Instead of getting an immediate “failed to process” error It loaded for a bit then gave me a 504. That’s the error that appears in the screenshot.
When I test it now, I don’t even get some loading and the 504. It went back to an immediate failed response. This is testing with o4-mini-deep-research and the default temperature field.
Thank you for your continued help!
+11Always happy to help,
I found this post on the OpenAI community that seems to imply that temperature isn’t supported with the o3 model, so I suspect that’s likely the reason behind the bug.
I tested with the o4-mini-deep-research model and included a temperature value just now and was able to replicate the same error. So I had a look in my OpenAI account to check I had enough credits and it turns out I’d run out. I bought some more then tested again and it was then able to run without hitting that same error.
Can you head over to the billing overview page in your OpenAI account and make sure it has enough API credits to perform the action?
Keep us posted on how it goes, hopefully it’s just a case of missing credits and not another bug! 🤞
Did you ever find a solution to this? I've been trying like crazy to get this to work and I keep running into the time out errors or the unable to process errors.
It seems like if it doesn't get a full response in a certain amount of time it just gives the time out error it's all I can think of
And it's interesting that when you select the perplexity deep research model it actually gives a warning that it's going to fail in testing which just confuses me because you can't really do anything then.
This post has been edited by a moderator to remove Zap link details. Please remember that this is a public forum and avoid any sharing potentially sensitive details.
+11Hi
The “unable to process request” errors that Cade was running into are due to a bug specifically affecting the use of the o3 model and setting a temperature. Can’t make any promises as to when that will be fixed but I’ve added you to the bug report to make sure you get an email notification from us as soon as it’s sorted.
The 504 error they also ran into, which I was able to replicate (with the o4-mini-deep-research model) seemed to have been due to a lack of API credits, so you’ll want to double-check there’s enough credits in your OpenAI account to handle the request.
With the “You have selected the Sonar Deep Research model, this is likely to fail if on testing step” notice seems to be referring to testing the the ChatGPT action within the Zap editor itself. If you switch the Zap on and run a live test does it timeout or return a similar “unable to process request” error?
+11Hey all 👋
I just came across a potential solution in a related topic and wanted to surface that here:
Fede2001 wrote:
Ok, sovled guys. I just had to verify my ID to access o3 for Api use. problem solved
Can you try verifying your ID in OpenAI to see if that then allows you access to use the o3 model with a temperature set?
Let me know whether that does the trick!
I’m reaching out regarding the use of the o3-deepresearch model in Zapier, and I wanted to highlight a few important points:
Required setup:
ID verification in OpenAI is required to access the model.
The Web Search tool must be enabled in the Zap configuration, or the model won’t run.
Timeout issue:
When using the deep research model, processing can take longer than 50 seconds, which often leads to timeout errors.
From what I understand, this may be due to Zapier’s internal timeout setting (e.g., OpenAI(timeout=50)).
It would be very helpful if you could increase the timeout limit, especially for long-running deep research tasks that naturally require more time to process.
Token limit issue:
The max_tokens value must be large enough to cover the full cost of input + reasoning + output.
If the limit is too low, the model may return an empty output.text field, even if it successfully completes the task.
I hope this feedback is helpful, and I’d really appreciate any update on whether timeout limits can be adjusted.
Thanks so much!
May be a good idea to automatically promote integrations in the future, but thats out of my forte. It’s odd that all the other models were working with the old integration, but deep research models were not.
+11I’m so sorry for the delay in my reply here,
Hi
I tested this further:
In test mode there was indeed a strict 50-second limit, but in live mode that limit didn’t apply.
However, when I tried using o4-mini-deep-research, I ran into a different issue. Even with a very simple prompt (just asking it to “find 5 interesting things”), I got a rate-limit error:
Rate limit reached for o4-mini-deep-research in organization org-xxxx on tokens per min (TPM): Limit 200000, Used 179288, Requested 25935. Please try again in 1.566s.
So even though the instruction was simple, the TPM usage spikes a lot.
Is that expected behavior for the o4-mini-deep-research model, or does this suggest something unusual in how it handles requests?
It seems the issue was caused by setting max_tokens to 200,000.
When max_tokens is very high, the model may try to allocate/output more tokens, which can easily push usage over the TPM limit.
On the other hand, if max_tokens is too low (e.g. 2,000–6,000), the model might not have enough tokens available to generate the response, which sometimes leads to an empty output.
What would be the recommended approach in this case?
+11Good questions,
“Rate limit reached for o4-mini-deep-research in organization org-xxxx on tokens per min (TPM): Limit 200000, Used 179288, Requested 25935. Please try again in 1.566s.” is a rate limit error. It’s saying that during the rolling 60-second window that you tested the ChatGPT action, your account had already used 179,288 tokens and performing that test would have used another 25,935 tokens. This would have taken it over the 200,000 token limit and so it returned that rate limit error.
According to OpenAI’s documentation on their rate limits it’s best to set the max_tokens to be as close as possible to the expected response size:
If you’re testing with that simple “find 5 interesting things” prompt try starting with max_tokens set to around 10,000-15,000 and see if that goes through successfully. You’ll likely need to adjust the amount when testing with a more complex prompt. Let us know how it goes!
Enter your E-mail address. We'll send you an e-mail with instructions to reset your password.
