When the AI foundry adopt GPT-5.2 latest version (40% faster)?

Question

I deployed GPT-5.2 model in Azure AI Foundry, but I found the GPT-5.2 is so slow that the response time could be greater than 2 minutes
The OpenAI Developers official account said they have optimized the model and now it's 40% faster, I'm wondering when the Azure OpenAI Service adopt this change?
https://x.com/OpenAIDevs/status/2018838297221726482

Answer 1

hi Jiaqi Zhang,

try review and reduce unnecessary context in your prompts, and if possible test another region. If latency is business critical, you may need to temporarily use a different model option or the public OpenAI API, provided that fits your compliance requirements. 40 percent faster announcement does not imply an immediate speed up in Azure. Azure OpenAI updates are delayed and rolled out gradually. There is no announced date for when this optimisation will reach Azure AI Foundry.

rgds,

Alex

Answer 2

Hello Jiaqi Zhang,

Welcome to Microsoft Q&A and Thank you for reaching out.

The OpenAI performance improvement does not get adopted by Azure OpenAI immediately. Azure rolls out model optimizations on its own schedule, after additional validation, compliance checks, and regional deployment.

What the OpenAI announcement actually means

The post from @OpenAIDevs is referring to:

OpenAI-hosted API infrastructure

Optimizations in their inference stack

Same model weights, but faster runtime execution

This improvement applies first to:

api.openai.com

OpenAI-managed endpoints

It does not automatically or instantly apply to Azure OpenAI / Azure AI Foundry.

How Azure OpenAI adopts model improvements

Azure OpenAI is a separate managed service that:

Runs models on Azure-owned infrastructure

Has additional layers for:

Security & compliance

  Capacity management
  
     Regional isolation
     
        Enterprise SLAs
        

Because of this, model optimizations are onboarded gradually, and only after:

Internal validation by Azure

Performance and stability testing

Region-by-region rollout

SKU / deployment compatibility checks

There is no public ETA announced for when a specific OpenAI optimization lands in Azure.

Why you’re seeing >2-minute latency

From the graph you shared (“Time to last byte”), the spikes strongly suggest a combination of:

High regional load

Queueing under rate limits

Non-streaming responses

Possibly Data Zone Standard or constrained regional capacity

This is not unusual for newly released, high-demand models like GPT-5.2 in Azure, especially shortly after public announcements.

Clarification

Deploying GPT-5.2 in Azure AI Foundry does not guarantee parity with OpenAI API latency, even if the model name and version are the same.

Same model ≠ same runtime stack.

What you can do today to reduce latency

While waiting for Azure to adopt the optimization:

Enable streaming responses (biggest UX win)

Reduce max_tokens where possible

Lower concurrency spikes (smooth traffic)

Check region capacity and test another supported region

Monitor Time to First Token (TTFT) vs Time to Last Byte

Watch for throttling in Azure Metrics / Logs.

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

Thank you!