Session: Token Economics and Model Cost Calculation

Full write-up coming soon.

The agent eval harness and OCR eval produced real cost data across five models on two task types — infrastructure diagnosis and vision OCR. That data is the foundation for a proper cost calculator.

Topics to be covered:

• Task type classification: infrastructure diagnosis vs OCR vs retrieval vs general reasoning
• Per-model cost curves based on measured data (not vendor estimates)
• Blended cost calculator for multi-model workflows (e.g., Haiku for triage + Sonnet for escalation)
• CLI design: task type + complexity → projected cost across models with routing recommendation