models.yml

providers:
  strix-halo:
    baseUrl: http://127.0.0.1:8081/v1
    api: openai-completions
    auth: none
    models:
      - id: Q38FN # server ignores the model field; canonical id is halogen-qwen3.8-flash-next
        name: Qwen 3.8 Flash Next
        api: openai-completions
        reasoning: true
        input: [text, image]
        tokenizer: qwen3
        contextWindow: 262144
        maxTokens: 16384 # caps thinking length: the endpoint stops thinking ~1k tokens before this
        compat:
          thinkingFormat: qwen-chat-template
          qwenPreserveThinking: true
          replayReasoningContent: true # required, or there is no reasoning in history for preserve_thinking to keep
          supportsDeveloperRole: false
          supportsStore: false
          supportsUsageInStreaming: true
          supportsForcedToolChoice: true
          supportsReasoningEffort: true # keeps Q38FN:xhigh role suffixes valid; no effort reaches the wire today
          reasoningEffortMap:
            minimal: minimal
            low: low
            medium: medium
            high: high
            xhigh: xhigh
            max: xhigh