hunjing/config/models.example.yaml

# 魂镜 AI Provider 模型配置
# 提供商基础设施配置移到了 config/providers.yaml
version: 1

models:
  # ==================== DeepSeek ====================

  - name: deepseek-v4-flash
    display_name: DeepSeek V4 Flash
    provider: deepseek
    model: deepseek-v4-flash
    format: openai
    tags: [default]
    supports_tools: true
    supports_structured_output: true
    supports_streaming: true
    supports_thinking: true
    supports_reasoning_effort: true
    context_window: 1000000
    response_reserve: 8192
    max_output_tokens: 384000
    timeout_seconds: 300
    routing_strategy: priority_failover
    endpoint_pool_enabled: true
    endpoints:
      - name: primary
        model: deepseek-v4-flash
        priority: 10
        weight: 100
        rpm_limit: 2500
      - name: fallback
        model: deepseek-v4-pro
        priority: 20
        weight: 50

  - name: deepseek-v4-pro
    display_name: DeepSeek V4 Pro
    provider: deepseek
    model: deepseek-v4-pro
    format: openai
    tags: [fallback]
    supports_tools: true
    supports_structured_output: true
    supports_streaming: true
    supports_thinking: true
    supports_reasoning_effort: true
    context_window: 1000000
    response_reserve: 8192
    max_output_tokens: 384000
    timeout_seconds: 300
    routing_strategy: priority_failover
    endpoint_pool_enabled: true
    endpoints:
      - name: primary
        model: deepseek-v4-pro
        priority: 10
        weight: 100

  - name: deepseek-v4-pro-anthropic
    display_name: DeepSeek Pro（Anthropic 格式）
    provider: deepseek
    model: deepseek-v4-pro
    format: anthropic
    supports_tools: true
    supports_structured_output: true
    supports_streaming: true
    supports_thinking: true
    supports_reasoning_effort: true
    context_window: 1000000
    response_reserve: 8192
    max_output_tokens: 384000
    timeout_seconds: 300
    endpoints:
      - name: primary
        model: deepseek-v4-pro
        base_url_env: DEEPSEEK_ANTHROPIC_BASE_URL
        priority: 10
        weight: 100

  # ==================== Anthropic / Claude ====================

  - name: claude-sonnet
    display_name: Claude Sonnet
    provider: anthropic
    model: claude-sonnet-4-6
    format: anthropic
    supports_tools: true
    supports_structured_output: true
    supports_streaming: true
    supports_thinking: true
    supports_reasoning_effort: true
    supports_vision: true
    context_window: 1000000
    response_reserve: 16000
    max_output_tokens: 16000
    timeout_seconds: 60
    reasoning_effort: medium

  - name: claude-haiku
    display_name: Claude Haiku
    provider: anthropic
    model: claude-haiku-4-5
    format: anthropic
    supports_tools: true
    supports_structured_output: true
    supports_streaming: true
    supports_vision: true
    context_window: 200000
    response_reserve: 4096
    max_output_tokens: 4096
    timeout_seconds: 60

  - name: claude-fable
    display_name: Claude Fable 5
    provider: anthropic
    model: claude-fable-5
    format: anthropic
    supports_tools: true
    supports_structured_output: true
    supports_streaming: true
    supports_thinking: true
    supports_reasoning_effort: true
    supports_vision: true
    context_window: 1000000
    response_reserve: 16000
    max_output_tokens: 16000
    timeout_seconds: 120
    reasoning_effort: high

  # ==================== Embedding ====================

  # 本地元炁推理引擎（默认，零配置可用）
  - name: bge-m3
    display_name: BGE-M3 Embedding（元炁本地）
    provider: yuanqi
    model: bge-m3
    format: openai
    tags: [embedding]
    timeout_seconds: 30
    dimensions: 1024
    endpoints:
      - name: primary
        model: bge-m3
        base_url_env: YUANQI_BASE_URL
        api_key_env: YUANQI_API_KEY
        priority: 10
        weight: 100

  # 云端备选（配好 API_KEY 后改为默认）
  # - name: text-embedding-3-small
  #   display_name: Text Embedding 3 Small
  #   provider: deepseek
  #   model: text-embedding-3-small
  #   format: openai
  #   tags: [embedding]
  #   timeout_seconds: 60
  #   dimensions: 1536

  # ==================== 小米 MiMo ====================

  - name: mimo-v2.5-pro
    display_name: MiMo v2.5 Pro
    provider: xiaomi
    model: mimo-v2.5-pro
    format: openai
    supports_tools: true
    supports_structured_output: true
    supports_streaming: true
    supports_reasoning_effort: true
    supports_vision: true
    context_window: 128000
    response_reserve: 8192
    max_output_tokens: 32768
    timeout_seconds: 300
    routing_strategy: priority_failover
    endpoint_pool_enabled: true
    endpoints:
      - name: primary
        model: mimo-v2.5-pro
        priority: 10
        weight: 100

  - name: mimo-v2.5
    display_name: MiMo v2.5
    provider: xiaomi
    model: mimo-v2.5
    format: openai
    supports_tools: true
    supports_structured_output: true
    supports_streaming: true
    supports_vision: true
    context_window: 128000
    response_reserve: 4096
    max_output_tokens: 16384
    timeout_seconds: 180

  - name: mimo-v2.5-asr
    display_name: MiMo v2.5 ASR
    provider: xiaomi
    model: mimo-v2.5-asr
    format: openai
    tags: [asr]
    context_window: 0
    timeout_seconds: 120
    endpoints:
      - name: primary
        model: mimo-v2.5-asr
        priority: 10
        weight: 100

  - name: mimo-v2.5-tts
    display_name: MiMo v2.5 TTS
    provider: xiaomi
    model: mimo-v2.5-tts
    format: openai
    tags: [tts]
    supports_streaming: true
    context_window: 0
    timeout_seconds: 120

  - name: mimo-v2.5-tts-voiceclone
    display_name: MiMo v2.5 TTS 声纹克隆
    provider: xiaomi
    model: mimo-v2.5-tts-voiceclone
    format: openai
    tags: [tts, voice_clone]
    supports_streaming: true
    context_window: 0
    timeout_seconds: 300

  - name: mimo-v2.5-tts-voicedesign
    display_name: MiMo v2.5 TTS 音色设计
    provider: xiaomi
    model: mimo-v2.5-tts-voicedesign
    format: openai
    tags: [tts, voice_design]
    supports_streaming: true
    context_window: 0
    timeout_seconds: 300