docs: Document dynamic model fetching feature and new model support (GPT-5.4, Gemma 4)

[continue]

Summary

PR #12046 introduced a new feature for dynamically fetching provider models that needs documentation.

Features to Document

1. Dynamic Model Fetching

• Users can now refresh/fetch available models dynamically from providers using their API key
• A refresh button appears in the Add Model UI after entering an API key
• Supports auto-loading models from:
  • Ollama: Scrapes the library with icons
  • OpenRouter: Fetches from public API on load
  • Anthropic: Uses /v1/models endpoint
  • Gemini: Uses v1beta/models (chat-only)
  • Other providers: Uses listModels() with API key

2. Auto-detection Features

• Context length, max tokens, and tool/image capabilities are automatically detected
• Results are deduplicated with hardcoded models (hardcoded take precedence)
• Proper provider icons displayed

3. Configuration Persistence

• contextLength, completionOptions.maxTokens, and capabilities (tool_use, image_input) are written to config.yaml when adding a model

4. New Model Support

• GPT-5.4: pro and mini variants
• Gemma 4: With improved image/tool detection

Suggested Documentation Locations

• Update /customize/model-providers/overview.mdx with information about dynamic model fetching
• Update individual provider pages (Anthropic, Gemini, OpenAI, etc.) as needed
• Consider adding a section in the Model Capabilities documentation

Related PR

• feat: fetch provider models dynamically and fix isOpenSource pollution #12046

---

Generated with Continue

Co-Authored-By: Continue noreply@continue.dev