o3-mini thread runs "forever"

Question

o3-mini thread runs "forever"

Andreas Gnyp 10

I've deployed o3-mini model and after some hickups, it was working fine. Now, since several days, a started run/thread stays in "running"-mode forever. There is not even a time-out. No error-message, no hint on what's wrong.

Manas R Mohanty 17,270 Reputation points Moderator

2025-09-26T07:54:44.29+00:00

Hi Andreas Gnyp

Could you share any share the region/ deployment type/Sample prompt (size too)/TPM/Response time or any specific replicable code for the same.

Replication

Below are test results in East Us2 region

"What are best practices to survive a high heat weather in 300 words". It took around 25 second for generating the response (Reasoning High)

What are best practices to survive a highly chill weather in 300 words. It took around less than 8 second (Reasoning low)

Reason

Slowness can occur if you are exceeding in payload with available TPM limit frequently (Reduces availability for Pay as you go models eventually auto triggered blocks) or there is a complexity in the prompts.

or There is backend region specific issue itself.

Looking forward to hearing from you.

Thank you.
Manas R Mohanty 17,270 Reputation points Moderator

2025-09-29T02:45:50.3366667+00:00

Hi Andreas Gnyp

We are still waiting to hear you confirmation on region/deployment type/prompt size.

Thank you
Andreas Gnyp 10 Reputation points

2025-09-29T15:28:53.03+00:00

Hi Manas Mohanty,

yes, and gladly:

Deployment info

[REDACTED here and moved to private message]

The rate limits are not exceeded. We're prompting with a maximum of ca. 10k tokens per request. Only with parallelization we'd go higher, but we're not doing it, yet. And it all worked just fine, now suddenly this.

What else do you need?

Your answer

Manas R Mohanty 17,270 Reputation points Moderator

2025-09-26T07:54:44.29+00:00

Hi Andreas Gnyp

Could you share any share the region/ deployment type/Sample prompt (size too)/TPM/Response time or any specific replicable code for the same.

Replication

Below are test results in East Us2 region

"What are best practices to survive a high heat weather in 300 words". It took around 25 second for generating the response (Reasoning High)

What are best practices to survive a highly chill weather in 300 words. It took around less than 8 second (Reasoning low)

Reason

Slowness can occur if you are exceeding in payload with available TPM limit frequently (Reduces availability for Pay as you go models eventually auto triggered blocks) or there is a complexity in the prompts.

or There is backend region specific issue itself.

Looking forward to hearing from you.

Thank you.
Manas R Mohanty 17,270 Reputation points Moderator

2025-09-29T02:45:50.3366667+00:00

Hi Andreas Gnyp

We are still waiting to hear you confirmation on region/deployment type/prompt size.

Thank you
Andreas Gnyp 10 Reputation points

2025-09-29T15:28:53.03+00:00

Hi Manas Mohanty,

yes, and gladly:

Deployment info

[REDACTED here and moved to private message]

The rate limits are not exceeded. We're prompting with a maximum of ca. 10k tokens per request. Only with parallelization we'd go higher, but we're not doing it, yet. And it all worked just fine, now suddenly this.

What else do you need?