Route under a hard token budget

01 / THE BRIEF

Your mission

Assign either small or large to each task. Maximize the sum of quality scores while staying within 900 total tokens. Each task must have one assignment. Return an object keyed by task ID. This is a fixed optimization fixture, not real provider performance.

Acceptance criteria

1. Assign every task exactly once.
2. Only use small or large model labels.
3. Keep the total at or below 900 tokens.
4. Maximize total fixture quality (25).

Input fixture

{
  "budget": 900,
  "tasks": [
    {
      "id": "summarize",
      "small": {
        "tokens": 100,
        "quality": 6
      },
      "large": {
        "tokens": 400,
        "quality": 9
      }
    },
    {
      "id": "reason",
      "small": {
        "tokens": 200,
        "quality": 3
      },
      "large": {
        "tokens": 500,
        "quality": 10
      }
    },
    {
      "id": "classify",
      "small": {
        "tokens": 100,
        "quality": 8
      },
      "large": {
        "tokens": 300,
        "quality": 9
      }
    }
  ]
}

Expected output shape

{"summarize":"small", "reason":"small", "classify":"small"}

This evaluates one visible fixture, not arbitrary code or a hidden generalization suite. Use any model outside this site. Don’t paste secrets or personal data.