o3-mini (Assistants API) rejects default temperature parameter in run creation

[aliakseikislou]

Confirm this is a Node library issue and not an underlying OpenAI API issue

• This is an issue with the Node library

Describe the bug

Description

The OpenAI Node.js client merges the default temperature setting into all API requests, including Assistants API run creations. However, the o3-mini reasoning model does not support sampling parameters like temperature, causing any inclusion of this parameter to result in a 400 error: “Unsupported parameter: 'temperature' is not supported with this model.”

Expected Behavior

The SDK should automatically omit unsupported sampling parameters for reasoning models like o3-mini, allowing the run creation to succeed without errors.

Actual Behavior

The client continues to merge the temperature default into the run creation request payload, and the Assistants API rejects it with a 400 error.

Notes

The error was previously reporting in #1318, but no further action was performed to fix it, and people are still getting it (me included).

To Reproduce

1. Instantiate the client with a default temperature:

const client = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY,
  temperature: 1
});

2. Create an assistant and a thread, then attempt to create a run with o3-mini:

await client.beta.threads.runs.create(thread.id, {
  model: 'o3-mini',
  reasoning_effort: 'medium',
  instructions: 'Some instructions.',
  assistant_id: assistantId,
  stream: true,
});

3. Observe the 400 error:

400 Unsupported parameter: 'temperature' is not supported with this model.

Code snippets

OS

Ubuntu 25.04

Node version

Node.js 22.14.0

Library version

openai 4.95.1

[RobertCraigie]

The OpenAI Node.js client merges the default temperature setting into all API requests, including Assistants API run creations

This is not true, we send only what you pass to the method.

This sounds like an issue with the underlying OpenAI API and not the SDK, so I'm going to go ahead and close this issue.

Would you mind reposting at community.openai.com?

[aliakseikislou]

@RobertCraigie is there any way to skip temperature at all when creating an OpenAI client? It seems the issue can still be with this parameter being set explicitly on an API client creation.

[RobertCraigie]

The SDK does not support such a parameter when instantiating the client.

openai-node/src/index.ts

Line 153 in 37ab638

export interface ClientOptions {

[aliakseikislou]

Sorry, yes, the client doesn't support such a parameter. It seems the temperature is being set somewhere internally and implicitly (when creating an assistant, probably). I'll head to community.openai.com for this.