[WH-2099]

Summary

Closes #3917.

Adds GPT-6 Astra, GPT-6 Sol, and GPT-6 Luna, with their official context/output limits, reasoning levels, capabilities, and standard API prices.
GPT-6 uses max_completion_tokens with Chat Completions and requests encrypted reasoning through the existing Responses path.
Model-specific parameter handling removes unsupported sampling/logprob options when reasoning is enabled and rejects Astra none reasoning and unsupported Chat tool/reasoning combinations before sending a request.
GPT-6 Chat requests accept the neutral standard/auto reasoning defaults.

SDK transport and tool-loop replay fixes are tracked in #4033 and draft PR #4035.
This model registration uses published dependencies and can be reviewed independently of that work.

Release Notes

• Add GPT-6 Astra, Sol, and Luna with 1,050,000-token context windows and 128,000-token output limits.
• Support their reasoning levels, vision, structured output, streaming, and function tools through the supported API paths.
• Apply GPT-6 request-parameter restrictions and report unsupported Chat tool/reasoning combinations before calling the API.

Change Type

• LLM plugin

LLM Plugin Checklist

• Tool interaction flow (model-specific API restrictions)
• Other LLM functionality (reasoning parameters)
• New models / model parameter fixes

Version

• Bumped the top-level manifest.yaml and project versions to 1.0.7.
• Declared dify_plugin>=0.10.2,<0.11.0 and locked the published 0.10.2 release.
• OpenAI remains locked to 3.26.0, including the existing Decisions integration.

Testing

• Clean Python 3.12.13 environment matching the manifest: uv run --frozen --python 3.12 pytest --ignore=tests/live -q — 317 passed.
• Validated model schemas, plugin registration, model assets, and all three catalog entries.
• Real OpenAI SDK requests through HTTP mocks covered all three models, Responses and Chat, and streaming and blocking responses: 12 combinations passed.
• Checked tool/reasoning restrictions, sampling filters, encrypted-content requests, token limits, and structured-output serialization.
• Changed-file Ruff checks, uv lock --check --offline, and git diff --check passed.
• Dify CLI 0.6.10 packaged the committed plugin source successfully.
• Live API requests and a deployed Dify environment were not tested.

[Copilot]

Copilot review overview

🟢 Approval recommended

The implementation consistently applies the documented GPT-6 contracts and includes comprehensive regression coverage.

Review effort: Balanced
Findings: None

What changed in this PR

Adds GPT-6 Astra, Sol, and Luna support to the OpenAI plugin while updating dependencies and request handling.

Changes:

• Registers GPT-6 models with capabilities, limits, pricing, and documentation.
• Enforces GPT-6 reasoning, sampling, tool-calling, and token-parameter constraints.
• Updates plugin versions, dependencies, lockfile, and regression coverage.

File	Description
models/​openai/​README.md	Documents GPT-6 support and constraints.
models/​openai/​manifest.yaml	Bumps plugin version.
models/​openai/​pyproject.toml	Updates project and SDK versions.
models/​openai/​uv.lock	Locks refreshed dependencies.
models/​openai/​models/​common_openai.py	Normalizes GPT-6 parameters.
models/​openai/​models/​llm/​chat.py	Adds Chat-specific GPT-6 handling.
models/​openai/​models/​llm/​responses.py	Enables GPT-6 Responses reasoning support.
models/​openai/​models/​llm/​gpt-6-astra.yaml	Defines Astra metadata and pricing.
models/​openai/​models/​llm/​gpt-6-sol.yaml	Defines Sol metadata and pricing.
models/​openai/​models/​llm/​gpt-6-luna.yaml	Defines Luna metadata and pricing.
models/​openai/​models/​llm/​_position.yaml	Adds models to catalog ordering.
models/​openai/​tests/​test_chat.py	Tests GPT-6 completion-token routing.
models/​openai/​tests/​test_catalog.py	Validates GPT-6 catalog contracts.
models/​openai/​tests/​test_routing.py	Covers reasoning and tool constraints.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Copilot review overview

🔵 Needs a closer look

The new model-specific API behavior is externally dependent and was not validated with live requests.

Review effort: Balanced
Findings: None

[Copilot]

🟢 Approval recommended

The catalog, request handling, documentation, dependency metadata, and regression coverage are consistent with the stated GPT-6 requirements.

0 open findings

🧠 Review effort: Balanced