[Feature]: Refresh recent AI model support and preserve server-configured model IDs

[wyuc]

Problem or Motivation

OpenMAIC’s built-in model/provider registry can lag behind fast-moving model releases. Recent models such as GPT Image 2, Claude Opus 4.7, Qwen3.6, DeepSeek V4, Kimi K2.6, Tencent HY3, Xiaomi MiMo v2.5, and newer Grok variants should be easier to select in the UI and configure from server-side env/YAML.

There is also an extensibility gap: when server config provides *_MODELS with a model ID that is not already in the local registry, the current sync path can drop that model. This makes newly released models difficult to use before the app registry is updated.

Proposed Solution

Refresh the built-in provider/model lists:

• OpenAI: add gpt-5.4-pro; add an OpenAI image provider for gpt-image-2, gpt-image-2-2026-04-21, gpt-image-1.5, gpt-image-1, gpt-image-1-mini, and chatgpt-image-latest.
• Anthropic: add claude-opus-4-7.
• Qwen: add qwen3.6-max-preview, qwen3.6-plus, qwen3.6-plus-2026-04-02, qwen3.6-flash, qwen3.6-flash-2026-04-16, and qwen3.6-35b-a3b.
• Qwen Image: add qwen-image-2.0-pro, qwen-image-2.0-pro-2026-03-03, qwen-image-2.0, and qwen-image-2.0-2026-03-03.
• DeepSeek: add deepseek-v4-pro, deepseek-v4-pro[1m], and deepseek-v4-flash.
• Kimi: add kimi-k2.6.
• xAI/Grok: add canonical grok-4.20-reasoning, grok-4.20, and grok-4.20-multi-agent, while keeping existing beta IDs for compatibility.
• Tencent/Hunyuan: add a provider-level integration using prefixes such as TENCENT and TENCENT_HUNYUAN. Do not add HY3 as an env prefix because HY3 is a model name, not a provider name.
• Xiaomi: add a Xiaomi MiMo provider with mimo-v2.5-pro and mimo-v2.5.

Improve server-config extensibility:

• Preserve unknown server-provided model IDs from *_MODELS as synthetic { id, name } entries instead of filtering them out.
• Keep aggregator-specific IDs out of default direct-provider lists unless they are official direct API IDs.
• Allow those IDs through *_MODELS or custom providers.

Update settings UI, i18n, .env.example, README docs, and targeted tests.

Related Issues / Routing

• Related to [Bug]: in provider-keys.md 1.Lowest-Friction Setup, DEFAULT_MODEL is gpt-4o-mini, but set ANTHROPIC_API_KEY #433, but does not close it. That issue is about provider key documentation and default model setup.
• Related to feat(config): add GLM-5.1 and GLM-5V-Turbo to GLM preset models #436, which already added GLM-5.1 and GLM-5V-Turbo. This issue should verify GLM coverage rather than duplicate that work.
• Add regression coverage around the server sync behavior previously tracked in bug: stale modelId/providerId after server provider sync #226.
• Does not cover Ollama OpenAI-compatible runtime issues from Bug: Using Ollama OpenAI-compatible endpoint causes "LLM returned empty response" #271 or Ollama visibility behavior from [Bug]: Ollama provider always visible in model selector even without any configuration #419.

Alternatives Considered

• Only document custom provider setup for new models. This avoids registry updates but makes common recent models harder to discover.
• Add all observed aggregator IDs directly to built-in providers. This is convenient but risks mixing official direct API IDs with OpenRouter or third-party routing IDs.
• Add a generic live model catalog fetch feature. This would be broader but is more complex than the immediate registry/config refresh.

Area

API / Backend

Additional Context

References:

• OpenAI GPT Image 2: https://developers.openai.com/api/docs/models/gpt-image-2
• GPT-5.5 API status: https://help.openai.com/en/articles/11909943-gpt-53-and-gpt-54-in-chatgpt
• Claude Opus 4.7: https://www.anthropic.com/news/claude-opus-4-7
• Qwen3.6 model list: https://help.aliyun.com/zh/model-studio/text-generation-model
• Qwen Image API: https://help.aliyun.com/zh/model-studio/qwen-image-api
• DeepSeek V4: https://api-docs.deepseek.com/zh-cn/guides/coding_agents
• Kimi K2.6: https://platform.moonshot.cn/
• Tencent HY3: https://huggingface.co/tencent/Hy3-preview
• xAI model docs: https://docs.x.ai/developers/models