feat(openai/azure): enable temperature when reasoning_effort=none for…

[henry-fung]

Summary

Starting from GPT-5.2 (confirmed also for GPT-5.1), when reasoning_effort is
set to "none", temperature and top_p become valid parameters. Passing them
with any other reasoning effort value causes a 400 API error. This PR implements
the conditional logic for gpt-5.1, gpt-5.2, and gpt-5.4 in both the OpenAI
and Azure OpenAI providers.

OpenAI provider

• gpt-5.1.yaml, gpt-5.2.yaml, gpt-5.4.yaml: add temperature to
  parameter_rules
• llm.py (_build_responses_api_params): strip temperature/top_p/logprobs
  when reasoning is active (non-"none")
• llm.py (_chat_generate): same stripping for the standard Chat Completions path

Azure OpenAI provider

• constants.py: add temperature ParameterRule to gpt-5.1, gpt-5.2,
  gpt-5.4 entries
• llm.py (_chat_generate_with_responses): restructure reasoning vs
  temperature/top_p as mutually exclusive — reasoning active → set
  reasoning: {effort} only; reasoning="none" or absent → pass
  temperature/top_p, skip reasoning param

Related Issues or Context

Implements the parameter constraints documented by OpenAI:

"The following parameters are only supported when using GPT-5.4 with reasoning
effort set to none: temperature, top_p, logprobs. Requests to GPT-5.4 or GPT-5.2
with any other reasoning effort setting … will raise an error."

Verified via direct API testing that GPT-5.1 follows the same rules.

This PR contains Changes to Non-Plugin

• Documentation
• Other

This PR contains Changes to Non-LLM Models Plugin

• I have Run Comprehensive Tests Relevant to My Changes

This PR contains Changes to LLM Models Plugin

• My Changes Affect Message Flow Handling (System Messages and User→Assistant Turn-Taking)
• My Changes Affect Tool Interaction Flow (Multi-Round Usage and Output Handling, for both Agent App and Agent Node)
• My Changes Affect Multimodal Input Handling (Images, PDFs, Audio, Video, etc.)
• My Changes Affect Multimodal Output Generation (Images, Audio, Video, etc.)
• My Changes Affect Structured Output Format (JSON, XML, etc.)
• My Changes Affect Token Consumption Metrics
• My Changes Affect Other LLM Functionalities (Reasoning Process, Grounding, Prompt Caching, etc.)
• Other Changes (Add New Models, Fix Model Parameters etc.)

gpt-5.1/5.2/5.4 — reasoning_effort vs temperature behavior:

reasoning_effort	temperature sent to API	Result
none (default)	✅ included	Success
low / medium / high	❌ stripped	Would cause 400 error

Version Control

• I have Bumped Up the Version in Manifest.yaml
  • models/openai/manifest.yaml: 0.3.4 → 0.3.5
  • models/azure_openai/manifest.yaml: 0.0.49 → 0.0.50

Dify Plugin SDK Version

• I have Ensured dify_plugin>=0.3.0,<0.6.0 is in requirements.txt

Environment Verification

Local Deployment Environment

• Dify Version is: , I have Tested My Changes on Local Deployment Dify

SaaS Environment

• I have Tested My Changes on cloud.dify.ai

[gemini-code-assist]

Code Review

This pull request introduces support for the reasoning_effort parameter in Azure OpenAI and OpenAI models, specifically targeting GPT-5 variants. It includes logic to conditionally include or strip parameters like temperature, top_p, and logprobs based on whether reasoning is active, as these are mutually exclusive in the underlying APIs. Review feedback highlights the need for consistent top_p parameter rules across model definitions and suggests ensuring logprobs is consistently stripped in the standard chat generation path when reasoning is enabled.

[gemini-code-assist]

The PR description states that top_p is also a valid parameter for GPT-5.1 when reasoning_effort is set to none. However, it is missing from the parameter_rules for the gpt-5.1 model entry here, while it was correctly added (or already present) for gpt-5.2 and gpt-5.4.

                ParameterRule(
                    name="temperature",
                    **PARAMETER_RULE_TEMPLATE[DefaultParameterName.TEMPERATURE],
                ),
                ParameterRule(
                    name="top_p",
                    **PARAMETER_RULE_TEMPLATE[DefaultParameterName.TOP_P],
                ),

[gemini-code-assist]

Similar to the Azure OpenAI constants, top_p should be added to the parameter rules for GPT-5.1 to allow users to configure it when reasoning is disabled.

  - name: temperature
    use_template: temperature
  - name: top_p
    use_template: top_p

[gemini-code-assist]

The PR description mentions that logprobs is also restricted when reasoning is active. While it is correctly stripped in _build_responses_api_params, it is missing from the stripping logic here in the standard _chat_generate path. This could lead to a 400 error if a user provides logprobs with a non-none reasoning effort.

Suggested change

	model_parameters.pop("temperature", None)
	model_parameters.pop("top_p", None)
	model_parameters.pop("temperature", None)
	model_parameters.pop("top_p", None)
	model_parameters.pop("logprobs", None)