Frequent errors when using o3-deep-research

API Bugs
1

In the last few days, I’ve been regularly getting the same error messages while trying to use o3:

An error occurred while processing your request. You can retry your request, or contact us through our help center at help<.>openai<.>com if the error persists. Please include the request ID wfr_0198f1f3f9397d3f99fca841cfd329dc in your message.

An error occurred while processing your request. You can retry your request, or contact us through our help center at help<.>openai<.>com if the error persists. Please include the request ID wfr_0198f2306e977a56a3da82d8aac88895 in your message.

An error occurred while processing your request. You can retry your request, or contact us through our help center at help<.>openai<.>com if the error persists. Please include the request ID wfr_019907f81cda7c13a50ed1f5f5668136 in your message.

An error occurred while processing your request. You can retry your request, or contact us through our help center at help<.>openai<.>com if the error persists. Please include the request ID wfr_01990aee3166783d9cd09498563494c8 in your message.

An error occurred while processing your request. You can retry your request, or contact us through our help center at help<.>openai<.>com if the error persists. Please include the request ID wfr_01990b9e44c07d68aaf1eeb05d034dac in your message.

My main hypothesis is that the model is running out of context while writing the report, but I haven’t included any arguments that would limit its context. All I’m doing is this:
resp = client.responses.create(
model=“o3-deep-research”,
input=prompt,
tools=[
{
“type”: “web_search_preview”,
},
],
background=True,
)

2

See if you have bizarre limits like this in your platform limits page:

— — — —
o3-deep-research 30,000,000 TPM 10,000 RPM 10,000,000 TPD
o3-deep-research-2025-06-26 250,000 TPM 3,000 RPM
o4-mini-deep-research 150,000,000 TPM 10,000 RPM 10,000,000 TPD
o4-mini-deep-research-2025-06-26 250,000 TPM 3,000 RPM

That’s what I have at tier-5 – less daily tokens than the amount per minute, and if the snapshot version were used (or switched to in internal follow-up calls, no daily number and the value is “default for all other models”.

That certainly could break stuff - as this model can use tokens per call into the millions.

BUG: It seems OpenAI might need to run all organizations though a model limit reset.

Then, the default max_output_tokens is not sufficient for success, you need to manually set it to 50k-100k (a setting which also will consume input rate limit per call).

It seems this model is not set up to be practical for use, or to get output once you have paid for internal grinding. $10.00 per million, and you only hope that the internal context-building gets cached. Means top-tier that can spend $100,000 can use this for just a few calls a day maxing out at about $100, and can service fewer daily requests than a single-user ChatGPT pro account.