payload:
    override:
      - models:
          - name: deepseek-v4-flash-0731
          - name: deepseek-v4-flash
        params:
          thinking.type: adaptive
          output_config.effort: max

    filter:
      - models:
          - name: deepseek-v4-flash-0731
          - name: deepseek-v4-flash
        params:
          - thinking.budget_tokens
```然后走 Claude 的

不走 v1/response

这样指定就行了