I’ve been using GPT-5 since Friday. Overall, I had thought it was faster than Sonnet most of the time. The gpt-5-fast model is in particular faster. However today, I seem to be having the exact opposite experience: all gpt-5 models are exceptionally slow. I’m using gpt-5 standard right now, and I just had it take over 15 minutes, to complete one prompt. Granted, I used this PRACT approach I’ve developed, so it was working through a story from Linear. However it is not the biggest story I’ve implemented this way, it was moderate in size. Every action the agent+llm took, however, were all VERY slow.
I don’t see any notifications right now about issues with any models, but the performance of GPT-5 has definitely PLUMMETED today. It is much slower than Sonnet usually was, and slower than I’ve seen by far since starting to use GPT-5 on Friday.
Am I the only one here (i.e. is it something local to me and my network?), or are other people having similar issues?
it depends on the task you’ve given it in my experience. GPT-5 generally tends to keep thinking until it’s more or less sure about a solution, and then it thinks some more.
GPT-5-low has some pretty amazing results but doesn’t play philosopher every time you give it a task.
I tend to switch between the 2 depending on the complexity of the task.
I’m not quite talking about slow in that manner. I mean, the LLM itself, seemed excessively slow in all that it did. Not just thinking, but all the other things as well. Whereas it was quite fast since Friday, today (even still) seems to be significantly slower overall…rate of processing, rate of output token delivery, as the agent and llm iterate through things like reading files, grepping, etc. its ALL very slow.
I have 1.2gbps internet, and I’m clocking well over 900mbps when I test right now. So its not my network or internet.
I have noticed, and I don’t know if this is new, or just how its been since GPT-5 became available and I’m just noticing. However I’ll often get some output, then there will be a pause, then get some more output, then a pause. This wasn’t as evident last night, but today since things do seem a bit faster, these pauses are more noticable.
Sometimes they can last a little while.
Its not really an issue, just thought I’d note it.