Cost per Green Run: Seven Free-Tier Myths and a Ledger You Can Run

### Myth 1: Free tokens mean free iteration

Tokens cost money, even if they appear free. Every retry still costs something. To measure this, log each attempt, not just requests. The cost never disappears; it's just shifted from the invoice to time spent. Consider how many retries your tasks actually need; that's your true cost.

### Myth 2: Cost per request is the number that matters

Many believe average spend per call is the key metric. But a green run—when a task successfully completes—is what truly matters. Calculate cost per green run, not just per request. Averages and percentiles can be misleading. A workflow that burns 60% of tokens in retries, even if billed at a low rate, is not cheap. Track spend per shipped task, not just per call.

### Myth 3: A cheaper model is a cheaper task

Choosing a cheaper model based on token price alone is flawed. Consider both price and the number of attempts required for your specific task. A model half the price with three times the attempts is not half the cost—it's actually one and a half times more expensive. Factor in latency and actual performance before migrating.

### Myth 4: The provider dashboard is your source of truth

Dashboards show token usage but often aggregate by day or key, not by individual tasks. They can give a false sense of accuracy. Your run log is the true source of attribution. Always cross-check dashboard data with your own logs; discrepancies usually lie with the logs.

### Myth 5: A free server is a free build box

Free server access doesn't mean zero cost. Analyze exit codes and wall clock time for each run. Interruptions often manifest as non-zero exit codes or stalled timestamps. Every limit is a potential budget drain until measured. Tools like MonkeyCode illustrate how free resources require careful tracking of retries and time costs.

### Myth 6: Retries are noise

Retries may seem insignificant, but they can significantly impact costs. Calculate the failure tax: divide tokens spent on failed runs (red) by those on successful runs (green). Retries are a major cost line item, often overlooked. Treat retries as a regular expense, not just noise.

### Myth 7: The demo works, so production can move to the free tier

A successful demo does not guarantee production success. Production environments require continuous load testing and handling of limit changes. Demos lack real-world pressure. Free tiers are great for iteration and learning, but they cannot replace contracts that provide SLAs, audit trails, and guaranteed capacity. For production, invest in a proper contract with support and reliability guarantees.