GPT-6 Models

[samimiashkan]

Hi Everyone.. I can get this (.NET) working with almost all OpenAi models (GPT-4, GPT-5) but it doesn't work with GPT-6 .

Is there something missing?

Any comment is highly appreciated.

Ashkan.

[John Karam (johnkaram02)]

Could you post the exact exception (status code and message) and which GPT-6 variant you're using (gpt-6-sol, gpt-6-luna, ...)? GPT-6 failures with Agent Framework so far have been model or endpoint compatibility problems rather than something missing in the framework, and the error message tells you which one you've hit. These are the ones reported so far:

1. Azure AI Foundry, project endpoint, gpt-6-luna (.NET: [FoundryAgent][gpt-6-luna]: [Bug]: Invalid parameter: 'text.format' of type 'json_schema' is not supported with model version gpt-6-luna-2026-09-22 #8718): the call returns HTTP 500, or 400 with 'text.format' of type 'json_schema' is not supported with model version gpt-6-luna-2026-09-22 / 'reasoning.effort' is not supported with this model. In that thread, the same request to the resource endpoint (https://<resource>.services.ai.azure.com/openai/v1/...) worked while the project endpoint (.../api/projects/<project>/openai/v1/...) failed, and gpt-6-sol worked on both. It was a service-side problem: the reporter said it was working again before the issue was closed. If you still see it, switch to the resource endpoint or gpt-6-sol to confirm.

2. Sampling parameters: GPT-6 models are reasoning models, and several projects have hit 400s for sending temperature (e.g. 0) or top_p (see "gpt-6-luna fails: temperature 0 not supported" in other repos). If you set them through ChatOptions (including inside ChatClientAgentOptions or per run), remove them for GPT-6, and control reasoning through ChatOptions.Reasoning instead:

   var options = new ChatClientAgentRunOptions(new ChatOptions
   {
       // Temperature = 0,   <- remove for GPT-6
       // TopP = ...,        <- remove for GPT-6
   });

3. Hosted tools: Python: [Bug]: LocalShellTool's hosted type="shell" tool is not supported on gpt-6-luna-2026-09-22 #8914 reports the hosted shell tool isn't supported on gpt-6-luna. If your agent registers hosted tools (code interpreter, shell, file search), try once without them to see whether the model call itself works.

4. Azure OpenAI deployments: with Azure, the model/deployment string has to be your deployment name, not the model name. If you created a new deployment for GPT-6, make sure you're passing that deployment's name.

The quickest way to tell these apart is to make the same request with the plain OpenAI / Azure.AI.OpenAI client (no agent) and the same parameters. If that fails too, it's model/endpoint compatibility (1, 2 or 4). If only the agent fails, it's something the agent adds (tools, response format) and worth an issue here with the request body.

[samimiashkan]

Hi John.. Thanks for the response.. I solved my problem, which was actually only compilation warnings by putting the right #pragma statement before the code.

It seems that Agent Framework is still in some kind of experimental state and some functions create compile warnings about the same fact.

Regards,