Unable to use the Azure OpenAI provider with GPT-5 or o3

[kyrre]

Describe the bug

When I choose any of the gpt-5 or o3 model variants for either the assistant and or codegen, then I get this error message:

Error code: 400 - {'error': {'message': 'Unrecognized request argument supplied: reasoning_effort', 'type': 'invalid_request_error', 'param': None, 'code': None}}

Will you submit a PR?

• Yes

Environment

yrre.wahl.kongsgaard@VVY9G2TM74 deathcon-workshop % uv run marimo env
{
  "marimo": "0.15.5",
  "editable": false,
  "OS": "Darwin",
  "OS Version": "24.6.0",
  "Processor": "arm",
  "Python Version": "3.13.3",
  "Binaries": {
    "Browser": "140.0.7339.133",
    "Node": "v22.12.0"
  },
  "Dependencies": {
    "click": "8.2.1",
    "docutils": "0.22",
    "itsdangerous": "2.2.0",
    "jedi": "0.19.2",
    "markdown": "3.9",
    "narwhals": "2.5.0",
    "packaging": "25.0",
    "psutil": "7.0.0",
    "pygments": "2.19.2",
    "pymdown-extensions": "10.16.1",
    "pyyaml": "6.0.2",
    "starlette": "0.48.0",
    "tomlkit": "0.13.3",
    "typing-extensions": "4.15.0",
    "uvicorn": "0.35.0",
    "websockets": "15.0.1"
  },
  "Optional Dependencies": {
    "altair": "5.5.0",
    "anywidget": "0.9.18",
    "duckdb": "1.3.1",
    "ibis-framework": "10.8.0",
    "nbformat": "5.10.4",
    "openai": "1.107.3",
    "pandas": "2.3.2",
    "polars": "1.31.0",
    "pyarrow": "21.0.0",
    "loro": "1.6.0",
    "pytest": "8.4.2",
    "python-lsp-ruff": "2.2.2",
    "python-lsp-server": "1.13.1",
    "ruff": "0.13.0",
    "sqlglot": "27.14.0"
  },
  "Experimental Flags": {
    "mcp_docs": false,
    "chat_sidebar": true,
    "rtc_v2": true,
    "external_agents": true,
    "inline_ai_tooltip": false,
    "performant_table_charts": false,
    "multi_column": false,
    "rtc": false
  }
}

Code to reproduce

No response

[kyrre]

For context, the Terraform resources are like this:

resource "azurerm_cognitive_deployment" "gpt5" {
  name                 = "gpt-5"
  cognitive_account_id = azurerm_cognitive_account.ira.id

  rai_policy_name = "DisableEverything"

  model {
    format  = "OpenAI"
    name    = "gpt-5"
    version = "2025-08-07"
  }

  sku {
    name     = "DataZoneStandard"
    capacity = 1000
  }

  version_upgrade_option = "OnceNewDefaultVersionAvailable"
}
``

[mscolnick]

@kyrre does this work without Azure? (just hitting OpenAI directly)

[kyrre]

@kyrre does this work without Azure? (just hitting OpenAI directly)

I tried it now and it works with OpenAI directly.

[mscolnick]

@kyrre it also works for me (was just checking in case it was a version issue with the library). I'm not sure what the correct solution is since i don't know if this is a thing with just your Azure deployment or all Azure deployments.

[kyrre]

@kyrre it also works for me (was just checking in case it was a version issue with the library). I'm not sure what the correct solution is since i don't know if this is a thing with just your Azure deployment or all Azure deployments.

The only reference to this that I could find was this: https://learn.microsoft.com/en-us/answers/questions/5519548/does-gpt-5-via-azure-support-reasoning-effort-and

I'll investigate this a bit more this week. 🤔