Formerly known as Azure AI Services or Azure Cognitive Services is a unified collection of prebuilt AI capabilities within the Microsoft Foundry platform
gpt-5-nano extremely slow
I currently use gpt-4.1-nano, and decided to switch to gpt-5-nano, however the speed is extremely slow compared to 4.1 even though nano series are supposed to be low reasoning and high speed, it takes around 10 seconds to receive a response. And this is not only the initial "cold start" response but constantly.
Foundry Tools
-
Manas R Mohanty • 17,270 Reputation points • Moderator
2025-08-18T13:29:18.5666667+00:00 Hi Sarge
Please provide the region of your deployment and sample prompts to be in same page.
Queries like "How are things around mars today" are taking 10-12 seconds with High reasoning and 16K max tokens
Even though GPT-5-nano is nano version of model GPT-5, it is a newly onboard reasoning model
Will check with product group on expected latency and get back to you.
Thank you
-
Manas R Mohanty • 17,270 Reputation points • Moderator
2025-08-20T07:54:34.2533333+00:00 Hi Sarge
Good day. I tested GPT-5-nano model deployment in Canada East region today.
Responses were super-fast and under 3 seconds.
Please share your region and confirm if you are using virtual network or proxy by any chance.
Thank you
-
Manas R Mohanty • 17,270 Reputation points • Moderator
2025-08-21T02:37:00.02+00:00 Hi Sarge
We could not hear from you. Hope you are able to test GPT-5-nano model with normal latency now.
Thank you
-
Dace • 0 Reputation points
2025-08-25T20:25:15.9966667+00:00 Hi, I recently ran into the same issue.
4o-mini - ~3.5s
5-mini - ~11s5-nano - ~11-18s
I noticed that nano is significantly slower than 5-mini, which is even more unintuitive.
We're deployed in US east 2
Sample prompt is just from wikipedia + this initial prompt.
"
Ensure the response is in HTML as a single line string and not in markdown. Do not modify, add, or remove any
<span data-type=\"customNoteLink\" data-link-key=\"id\">elements. These<span>elements are critical for linking and should remain untouched in the response. You will fix spelling and grammar in the text in valid HTML format"Antarctica is Earth's southernmost and least-populated continent. Situated almost entirely south of the Antarctic Circle and surrounded by the Southern Ocean (also known as the Antarctic Ocean), it contains the geographic South Pole. Antarctica is the fifth-largest continent, being about 40% larger than Europe, and has an area of 14,200,000 km2 (5,500,000 sq mi). Most of Antarctica is covered by the Antarctic ice sheet, with an average thickness of 1.9 km (1.2 mi).
Antarctica is, on average, the coldest, driest, and windiest of the continents, and it has the highest average elevation. It is mainly a polar desert, with annual precipitation of over 200 mm (8 in) along the coast and far less inland. About 70% of the world's freshwater reserves are frozen in Antarctica, which, if melted, would raise global sea levels by almost 60 metres (200 ft). Antarctica holds the record for the lowest measured temperature on Earth, −89.2 °C (−128.6 °F). The coastal regions can reach temperatures over 10 °C (50 °F) in the summer. Native species of animals include mites, nematodes, penguins, seals and tardigrades. Where vegetation occurs, it is mostly in the form of lichen or moss.
"
-
-
Sarge • 65 Reputation points
2025-08-30T01:18:54.1866667+00:00 2 weeks since my question and still super slow.
-
-
Sarge • 65 Reputation points
2025-09-06T20:43:19.85+00:00 dude what kind of a solution suggestion is this? how about I change providers? heard aws bedrock is better.
-
Manas R Mohanty • 17,270 Reputation points • Moderator
2025-09-08T03:10:58.21+00:00 Hi Sarge
Good day.
We always recommend customer to load balance between multiple regions to reduce overall downtime.
Tested in East US 2 deployment today, Response time of greeting text is fine but taking more than 20 second for Reasoning queries
Please help with details requested in private message to create support ticket and escalate to product group.
Thank you
-
Allan Hu (胡雅伦) • 0 Reputation points
2026-01-06T07:18:25.6233333+00:00 I am facing the same issue now, it takes around 13 ~ 17 sec to response when I use azure's GPT-5-nano.
The resource is under eastUS2 region. I can not change to Canada, since I'm using the company's subscription/resources
Sign in to comment