I'll take a look at increasing the

[odd folio]

Unfortunately, the link I sent you only changes the timeout of the entire workflow and not the model's request.
It seems like the only way to increate timeout for an http request would be using the dedicated HTTP node but that wouldn't fit with a N8N AI Agent :/
https://community.n8n.io/t/n8n-time-out/12904

[plush jackal]

Yer :/

[plush jackal]

Theres an option for Timeouts under the OpenAI Chat model, I've increased it to five minutes for now

[odd folio]

Oh yeah you're right, I didn't see that

Does it work now ?

[plush jackal]

Running it to see if it does

5 minutes, should be plenty long enough really...

This was all working fine with GPT-4o before GPT-5 released, then GPT-4o went down the drain, gave inconsistent replies, was lazy and not looking at the input etc...

GPT-Turbo worked well, but that's crazy expensive...

[odd folio]

The fact that N8N doesn't support response streaming could be another reason of why it crashes here.

[plush jackal]

First run
Second run...
Seems the time increases per batch?

[odd folio]

Because instead of getting a few tokens regularly while the model is reasoning, it waits too long for a huge response, and close the connection

[plush jackal]

4687 tokens on the request

165945ms response time

[odd folio]

Well depending on which type of N8N Ai agent you're using it might be sending previous messages in the conversation to keep context, so it'd be somewhat logical that time increases

[plush jackal]

165 seconds, not sure how to convert that for token usage per second

[odd folio]

Both your requests are 4k long and one of them takes twice the time ?

[plush jackal]

Yes

There is no memory module attached, so it shouldn't be passing previous conversations through

[odd folio]

You're using GPT5 ? And are those 2 different messages or did you send the same twice to verify ?

If it's 2 different messages of different difficulties, GPT-5's internal router might send one to the standard version while the other prompt is router towards the gpt-5-thinking model, hence it takes more time ?

[plush jackal]

Tell a lie...

First request was 3016 - 75971ms
Second was 4687 - 165945ms
Third was 3937 - 139268ms
Fourth was 2810 - 55679ms
Fifth was 3856 - 123840
Sixth was 4382 - 70887ms

The response times, all vary quite significantly

[odd folio]

Yeah ok seems normal then ?
4.6k token's a long request, and depending on the task you're asking do you feel like 50-130 seconds is too long ?

[plush jackal]

All the data is relatively the same - It's preformatted...
And the prompt is the same.

I have twenty different tokens of data, of all the same size being passed through

[plush jackal]

Well, I can only compare to gpt-4o and GPT4-turbo - Where they were both exceedingly fast... No timeout errors or anything, run first time everytime...

[odd folio]

Yeah but 4o and Turbo aren't reasoning models

[plush jackal]

But switching to GPT-5 I am getting timeout errors

[odd folio]

They're inherently different

[plush jackal]

Well, yes 4o was for data analysis, which is why i was using it, but on release fo GPT-5 - It started giving inconsistent issues, and hallucinating

So I then used GPT4-turbo - But that's blown my months budget in a day...

[odd folio]

Just out of curiosity, wouldn't gpt5mini or gpt5nano already answer your prompts correctly ?

[plush jackal]

what do you mean?

[odd folio]

There's 4 publicly available GPT5 versions :

• GPT5 Chat
• GPT5 (normal)
• GPT5 Mini
• GPT5 Nano

From what I've understood, you use GPT5 (normal)

But it might be overkill if gpt4o was able to answer you correctly. Trying Mini or Nano versions could be interesting

[plush jackal]

Okay, would they be more for data analysis then?

[odd folio]

I couldn't tell you because I mainly use AI to code, but open-ai has given its view on what each gpt5 model could be compared to
https://openai.com/index/gpt-5-system-card/

If you're looking for a gpt4o equivalent, you might look into GPT5 without thinking, on n8n that would be gpt-5-chat-latest

[plush jackal]

Ah okay - I thought the chat - would just be chat prompts etc...

When I saw GPT-5-main - I assumed it was GPT-5

[odd folio]

Yeah it's kinda confusing

Using OpenRouter you can check that GPT-5 (normal) cannot have reasoning disabled but GPT-5 Chat can

I'm interested to know if changing from gpt-5 to gpt-5-chat reduces the overall processing time @plush jackal 🙂

[plush jackal]

So am I - I'm still waiting for this run to finish though 🤣

[odd folio]

From my tests, it should.
I simply sent :

• Hi
• Hi, think way harder
  I got 2.5sec and 2.5sec for gpt-5-chat
  Whereas I got 1.5sec and 11sec for gpt-5

That’s because of the internal router in gpt-5 that sends harder questions to a more capable model

[odd folio]

Fingers crossed 😂

[plush jackal]

So from the time differences - Between 10 and 20 (20 is max) inputs - The average is around 70,000ms with gpt-5. with two showing at over 100,000ms for response.

13 was the highest at 184,000ms
Whilst 12 was the lowest with only 43,000ms

Which is strange, they all have the same prompt, and the same amount of data but the fluctuations are quite different

I'll check gpt-5-chat shortly, for some reason it seems to take a couple hours, whenever I change the model on the API key

[odd folio]

Alright !!

[plush jackal]

I appreciate the advice and conversation, thank you

[odd folio]

Ofc !
I just tried it with some random data, ~6000 tokens. It timed out with gpt-5 (>2 min) and I got an answer in 24sec with gpt-5-chat

[plush jackal]

How was the response in comparison

[odd folio]

gpt-5 timed out because I didn’t specify a higher timeout parameter :/ I didn’t get a response

Did you try ?

[plush jackal]

Not with GPT-5-Chat no, It takes hours to sync the models to the API key apparantly...

When I add a new model

It shows here, almost immediately - But when I run it I get API Permission errors

[plush jackal]

20425ms response time using GPT-5-Chat

Way faster!

30k is the slowest so far

[odd folio]

Nice !

[odd folio]

Is the response satisfactory compared to gpt-5 ?

[plush jackal]

Will review it properly shortly, got collared by my daughter, I'll let you know

[plush jackal]

This was GPT-5's output

This was GPT-5-Chat output

The output on GPT-5-Chat is much better!

I do all the analysis and screening through code - Then I pass it to AI to find potential Entry / Exit conditions based on all my analysis, and to explain why it's chose that.

It's purpose is more of an advanced Guidance/Screener on top of my own, where AI can detect through Data, things I might miss...

Pretty neat 🙂

The price is also - Much much better!...

[plush jackal]

The price is almost astonishing...
From $1.20 per run to $0.11c per run, which brings it back to around GPT-4o levels