Azure OpenAI: Deploying fine-tuned gpt-4o (2024-08-06) with DeveloperTier SKU fails with InvalidResourceProperties

Ava 0 Reputation points
2026-10-22T17:07:07.07+00:00

Hello,

I fine-tuned gpt-4o (2024-08-06) using the Developer training tier in Sweden Central. I'd like to serve it with the Developer deployment type (per-token billing, no hourly hosting fee) for short, occasional evaluation sessions.

What I tried (Azure CLI in Cloud Shell):

az cognitiveservices account deployment create -n <resource-name> -g <resource-group> --deployment-name dev-test --model-format OpenAI --model-name gpt-4o-2024-08-06.ft-<id> --model-version 1 --sku-name DeveloperTier --sku-capacity 50

Result:

(InvalidResourceProperties) The specified SKU 'DeveloperTier' of account deployment is not supported by the model 'gpt-4o' version: '2024-08-06'.

Request ID: c7dc1f01-5804-49c0-9bc9-66e608961024 (Oct 22, 2026, ~16:33 UTC)

The fine-tuning overview lists gpt-4o as not supported for Developer deployment, and the portal only offers Global Standard, Standard and Provisioned Throughput. The Azure retail price list does include "gpt 4o dev ft" input/output token meters, which is why I'm asking.

Question: Is there any supported way to serve a fine-tuned gpt-4o (2024-08-06) model on the Developer deployment type, such as preview access for a subscription? Or is Global Standard (with hourly hosting) the only option?

Many thanks!

Update 10/22/26: I also tried --sku-name Developer (as the automated AI answer suggested) and got The deployment sku name 'Developer' is invalid. So DeveloperTier is the recognized name, but it's rejected for gpt-4o 2024-08-06.

Update 11/1/26: Still looking for help with this!

Azure OpenAI in Foundry Models
0 comments

1 answer

Sort by: Most helpful
Most helpful Newest Oldest
  1. Rukshan edirisinghe 2,365 Reputation points
    2026-10-23T03:53:54.1733333+00:00

    Hi @Ava

    You read the overview correctly, and the error confirms it. In the fine-tuning supported models table, gpt-4o (2024-08-06) has a footnote on serverless deployment that excludes the Developer deployment type. It can be trained on the Developer tier, but not served with Developer deployment. No CLI flag will change that.

    Two practical options for short evaluation sessions:

    1. Retrain on a model that supports Developer deployment. gpt-4.1, gpt-4.1-mini and gpt-4.1-nano support it, with per-token billing and no hourly hosting fee. A Developer deployment is for evaluation only, lasts 24 hours and has no SLA, which fits occasional testing.
    2. Keep gpt-4o and deploy it as Global Standard (or Standard in Sweden Central), then delete the deployment as soon as each evaluation session ends. That keeps hosting charges to the hours you actually use it. The fine-tuned model itself stays in your resource, so you can redeploy later.

    If gpt-4.1-mini gives similar quality on your data, option 1 is the cheapest way to keep testing.

    Is gpt-4o specifically required for your use case, or could you compare it with a gpt-4.1 fine-tune?

    If this unblocked your evaluation, please consider accepting it as the answer. If the Global Standard deployment also fails, let me know and I'll be happy to keep helping.

    Reference:

    Was this answer helpful?

    0 comments

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.