Context

Qwen/DashScope is currently usable through a custom OpenAI-compatible ModelConfig, but it is not first-class in yoagent:

• No ModelConfig::qwen(...) / ModelConfig::dashscope(...)
• No OpenAiCompat::qwen()
• ThinkingFormat::Qwen already exists, but no preset uses it
• The provider is absent from the Model Presets docs

This should be added as a first-class OpenAI-compatible preset, similar to DeepSeek, MiniMax, and Z.ai.

Current official API references

Primary OpenAI-compatible docs:

• Qwen Cloud OpenAI compatibility overview:
  https://docs.qwencloud.com/api-reference/toolkitframework/openai-compatible/overview
• Alibaba Cloud Model Studio Qwen API reference:
  https://www.alibabacloud.com/help/en/model-studio/qwen-api-reference/
• Alibaba Cloud OpenAI-compatible text generation examples:
  https://www.alibabacloud.com/help/en/model-studio/text-generation

Feature-specific docs:

• Streaming with OpenAI-compatible SSE and stream_options.include_usage:
  https://www.alibabacloud.com/help/en/model-studio/stream
• Function calling / tools:
  https://www.alibabacloud.com/help/en/model-studio/qwen-function-calling
• OpenAI Responses API for Qwen:
  https://www.alibabacloud.com/help/en/model-studio/qwen-api-via-openai-responses
• Model list / capability overview:
  https://www.alibabacloud.com/help/en/model-studio/models

Relevant API details from docs

OpenAI-compatible Chat Completions base URLs are region-specific:

• International/Singapore: https://dashscope-intl.aliyuncs.com/compatible-mode/v1
• China/Beijing: https://dashscope.aliyuncs.com/compatible-mode/v1
• US/Virginia: https://dashscope-us.aliyuncs.com/compatible-mode/v1

Auth:

• Authorization: Bearer $DASHSCOPE_API_KEY

Endpoint:

• POST /chat/completions

Examples in current docs use model IDs such as:

• qwen3.6-plus
• qwen3.5-plus
• qwen-plus
• qwen-flash
• qwen3-coder-plus
• qwen3-coder-flash

Qwen-specific Chat Completions parameters include:

• enable_thinking
• thinking_budget
• enable_search
• search_options
• top_k

Streaming is documented with:

"stream": true,
"stream_options": { "include_usage": true }

Tool calling is supported through OpenAI-compatible tools / tool_calls for supported Qwen text and selected multimodal models.

Proposed yoagent changes

Add an OpenAI-compatible Qwen/DashScope preset:

• Add OpenAiCompat::qwen().
• Add ModelConfig::qwen(id, name) using the international base URL by default:
  https://dashscope-intl.aliyuncs.com/compatible-mode/v1
• Optionally add region-specific helpers, or document how callers can override base_url for China/US regions.
• Set:
  • provider = "qwen" or "dashscope"; choose one and document it
  • api = ApiProtocol::OpenAiCompletions
  • compat = Some(OpenAiCompat::qwen())
  • thinking_format = ThinkingFormat::Qwen
  • supports_usage_in_streaming = true
  • max_tokens_field = MaxTokensField::MaxTokens
• Decide whether supports_reasoning_effort should remain false; current Qwen docs use Qwen-specific enable_thinking / thinking_budget, not OpenAI reasoning_effort.

Optional follow-up:

• Extend OpenAiCompat with Qwen-specific thinking controls if yoagent wants ThinkingLevel to map to enable_thinking / thinking_budget.
• Consider OpenAI Responses API support for Qwen separately. Qwen Responses uses a different base URL:
  https://dashscope-intl.aliyuncs.com/api/v2/apps/protocols/compatible-mode/v1

Docs/tests

Add tests for:

• OpenAiCompat::qwen() flags.
• ModelConfig::qwen(...) provider, base URL, protocol, and compat.
• Request body uses max_tokens, not max_completion_tokens.
• Existing reasoning_content parsing still works for Qwen streams.

Update docs:

• docs/providers/model-presets.md
• docs/providers/openai-compat.md
• docs/providers/overview.md
• README provider table if needed

Why this matters

Qwen/DashScope is a common hosted path for Qwen models and exposes an OpenAI-compatible API with tool calling and streaming. Since yoagent already supports adjacent first-class OpenAI-compatible providers, adding Qwen would reduce custom config boilerplate and make provider behavior explicit.