Azure OpenAI: Missing latest models, incomplete reasoning detection, stale API version placeholder

[infinitelyloopy-bt]

Summary

Several gaps in Azure OpenAI support prevent users from fully utilizing the latest Azure OpenAI models (o3, o4-mini, codex-mini, gpt-4.1-nano, gpt-5-pro) and cause parameter mismatches for reasoning models.

Issues Found

1. Missing Models in azureChatOpenAI model list

The following GA models available on Azure OpenAI are missing from models.json:

• o3 (GA 2025-04-16) — full reasoning model
• o3-pro (GA 2025-06-10) — high-compute reasoning
• o4-mini (GA 2025-04-16) — compact reasoning model
• codex-mini (GA 2025-05-16) — code-focused reasoning
• gpt-4.1-nano (GA 2025-04-14) — ultra-low-cost model
• gpt-5-pro (GA 2025-10-06) — high-compute GPT-5

2. Reasoning Model Detection Incomplete

Current detection in both AzureChatOpenAI.ts and ChatOpenAI.ts:

if (modelName.includes('o1') || modelName.includes('o3') || modelName.includes('gpt-5'))

Problems:

• Misses o4-mini and codex-mini (both are reasoning models)
• False positive on gpt-5.2-chat and gpt-5-chat-latest (non-reasoning chat variants)

3. maxCompletionTokens Conversion Missing in ChatOpenAI

AzureChatOpenAI.ts correctly converts maxTokens → maxCompletionTokens for reasoning models, but ChatOpenAI.ts does not. This causes API errors when using reasoning models (o1, o3, o4-mini, gpt-5+) with a max tokens value set, since these models reject the max_tokens parameter.

Related: #5804

4. Stale API Version Placeholder

The credential field azureOpenAIApiVersion shows 2023-06-01-preview as placeholder text. This is a deprecated version from over 2 years ago. Should show 2024-10-21 (current GA).

The documentation link also points to the deprecated cognitive-services path instead of the current ai-foundry path.

5. No Custom Model Name Input

Azure deployment names are user-defined and may not match model names. The model dropdown only allows selection from the predefined list, with no option to type a custom name.

Environment

• Flowise v3.0.13
• @langchain/openai v0.6.3
• openai v4.96.0

Expected Behavior

• All GA Azure OpenAI models available in the dropdown
• Reasoning model detection covers o4-mini, codex-mini, and excludes chat variants
• maxCompletionTokens used consistently for all reasoning models
• Current GA API version shown as placeholder
• Users can type custom deployment names

[HenryHengZJ]

WIP: https://github.com/FlowiseAI/Flowise/tree/chore/Langchain-V1