Slowness in Azure OpenAI Gpt4.1

Question

Slowness in Azure OpenAI Gpt4.1

Roni Mateless 40

"Experiencing significant slowness with gpt-4.1 responses in the East US2 region. I haven't seen any status updates regarding this. Could we please get an update on the current situation?"

Roni Mateless 40 Reputation points

2025-04-23T13:39:49.9566667+00:00

Thanks @SriLakshmi C for your response.

pls check the average response time at least between 7:00-9:00 AM 23/4 (GMT) compared to the usual response time

Unfortunately, those delays are not reported in the status page

Anyway, when switching back to Gpt4o (latest) the response time is much faster (gpt4.1 is only available in East US2 in US).
SriLakshmi C 5,130 Reputation points Microsoft External Staff Moderator

2025-04-24T14:44:06.72+00:00

Hi @Roni Mateless

I checked the average response time for the window between 7:00–9:00 AM GMT on April 23rd, and it appears there was an issue during that period. The metrics are not showing any data for that timeframe, as you can see in the attached screenshot. I also checked the average response time for the window between 11:00 AM–1:00 PM GMT on April 23rd, and the metrics are displaying as expected without any delays. Please find the screenshot attached for reference. Thank you!
Roni Mateless 40 Reputation points

2025-04-24T22:18:13.8233333+00:00

That's makes sense, thanks for checking. I'll switch back to gpt4.1
SriLakshmi C 5,130 Reputation points Microsoft External Staff Moderator

2025-04-25T09:37:42.09+00:00

Hi @Roni Mateless,

Switching back to GPT-4.1 sounds like a good plan if it's working more reliably for your use case. If you run into any other issues or have questions about performance or configurations, feel free to reach out happy to help anytime!

I’ve converted my earlier comment into an actual answer so please can you upvote and accept it if it helped. Appreciate it!
Roni Mateless 40 Reputation points

2025-04-25T14:09:03.8233333+00:00

After switching back to GPT-4.1 and monitoring response times, it appears that—at least for our use cases, which typically involve 1000–1500 completion tokens—GPT-4.1 is noticeably slower than GPT-4o. On average, responses take around 45 seconds compared to 20 seconds with 4o. It seems like the resource allocation for 4.1 might still be limited.
SriLakshmi C 5,130 Reputation points Microsoft External Staff Moderator

2025-04-28T14:50:00.2933333+00:00

Hi @Roni Mateless,

You're correct GPT-4.1 generally has a higher latency compared to GPT-4o, especially for workloads involving a larger number of completion tokens (like your 1000–1500 token range). GPT-4o was specifically optimized for faster response times and higher throughput, while GPT-4.1 maintains a focus on slightly different trade-offs, including model behavior and quality for certain tasks.

Additionally, resource allocation for GPT-4.1 can indeed vary depending on overall service demand and capacity management, which might contribute to the increased response time you're seeing.

While performance tuning is continuously being evaluated, for lower latency requirements, GPT-4o is currently the recommended model. If faster responses are critical for your scenario, continuing with GPT-4o would likely provide a better experience.

Thank you!

Accepted answer

0 additional answers

Your answer

Roni Mateless 40 Reputation points

2025-04-23T13:39:49.9566667+00:00

Thanks @SriLakshmi C for your response.

pls check the average response time at least between 7:00-9:00 AM 23/4 (GMT) compared to the usual response time

Unfortunately, those delays are not reported in the status page

Anyway, when switching back to Gpt4o (latest) the response time is much faster (gpt4.1 is only available in East US2 in US).
SriLakshmi C 5,130 Reputation points Microsoft External Staff Moderator

2025-04-24T14:44:06.72+00:00

Hi @Roni Mateless

I checked the average response time for the window between 7:00–9:00 AM GMT on April 23rd, and it appears there was an issue during that period. The metrics are not showing any data for that timeframe, as you can see in the attached screenshot. I also checked the average response time for the window between 11:00 AM–1:00 PM GMT on April 23rd, and the metrics are displaying as expected without any delays. Please find the screenshot attached for reference. Thank you!
Roni Mateless 40 Reputation points

2025-04-24T22:18:13.8233333+00:00

That's makes sense, thanks for checking. I'll switch back to gpt4.1
SriLakshmi C 5,130 Reputation points Microsoft External Staff Moderator

2025-04-25T09:37:42.09+00:00

Hi @Roni Mateless,

Switching back to GPT-4.1 sounds like a good plan if it's working more reliably for your use case. If you run into any other issues or have questions about performance or configurations, feel free to reach out happy to help anytime!

I’ve converted my earlier comment into an actual answer so please can you upvote and accept it if it helped. Appreciate it!
Roni Mateless 40 Reputation points

2025-04-25T14:09:03.8233333+00:00

After switching back to GPT-4.1 and monitoring response times, it appears that—at least for our use cases, which typically involve 1000–1500 completion tokens—GPT-4.1 is noticeably slower than GPT-4o. On average, responses take around 45 seconds compared to 20 seconds with 4o. It seems like the resource allocation for 4.1 might still be limited.
SriLakshmi C 5,130 Reputation points Microsoft External Staff Moderator

2025-04-28T14:50:00.2933333+00:00

Hi @Roni Mateless,

You're correct GPT-4.1 generally has a higher latency compared to GPT-4o, especially for workloads involving a larger number of completion tokens (like your 1000–1500 token range). GPT-4o was specifically optimized for faster response times and higher throughput, while GPT-4.1 maintains a focus on slightly different trade-offs, including model behavior and quality for certain tasks.

Additionally, resource allocation for GPT-4.1 can indeed vary depending on overall service demand and capacity management, which might contribute to the increased response time you're seeing.

While performance tuning is continuously being evaluated, for lower latency requirements, GPT-4o is currently the recommended model. If faster responses are critical for your scenario, continuing with GPT-4o would likely provide a better experience.

Thank you!

Answer 1

Hello @Roni Mateless,

I’ve attempted to reproduce the slowness with GPT-4.1 in the East US2 region in my own environment, but everything appears to be functioning normally without any noticeable latency.

However, I understand you’re experiencing delays, and this can sometimes occur due to factors such as high regional demand or usage limits. Here are a few suggestions to help troubleshoot and improve response times:

Sometimes, Azure may experience outages or high usage in a specific region. You can check the Azure status page for any ongoing issues.

Ensure that you're operating within your allocated service quotas. The usage limits for the GPT-4.1 model vary based on your subscription tier:

Enterprise Tier: Up to 5 million tokens per minute and 5,000 requests per minute
Default Tier: Up to 1 million tokens per minute and 1,000 requests per minute

GPT-4.1 supports a large context window, which can affect response times. If you're experiencing slowness, try reducing the size of your input prompts to see if it improves performance.

I Hope this helps. Do let me know if you have any further queries.

Thank you!

Share via

Slowness in Azure OpenAI Gpt4.1

0 additional answers

Your answer