paygo-ptu-cost-analyzer

To compare Azure OpenAI PAYGO versus PTU costs, you process per-minute token usage logs to compute percentile TPM capacity and estimate spillover-to-PAYGO scenarios. This analysis outputs markdown reports, JSON files, and PNG charts for capacity planning.

The required CSV token schema for PTU sizing includes minute_bin, input_tokens_sum, cached_tokens_sum, and output_tokens_sum columns. You must provide token usage data formatted with these specific fields to generate accurate percentile TPM envelope sizing and spillover estimates.

You can use Log Analytics AzureMetrics KQL queries instead of CSV exports for PAYGO vs PTU analysis. The headless report generator accepts either a customer CSV token export or a direct Log Analytics query to compute percentile-based cost and spillover reports.

PTU sizing analysis supports custom percentile selection for TPM capacity planning. You can specify percentile ranges such as p50 through p99 to model different capacity scenarios, evaluate PTU capacity selection, and estimate spillover-to-PAYGO costs for Azure OpenAI provisioning.

Generating a PAYGO vs PTU cost report produces report.md for markdown review, machine-readable report.json, and PNG charts for visual capacity planning. These outputs support review in pull requests or capacity-planning docs for Azure OpenAI provisioning decisions.

You need Azure identity credentials to query Log Analytics for token usage analysis when using AzureMetrics KQL queries instead of CSV exports. The analysis relies on azure-identity and azure-monitor-query dependencies to authenticate and retrieve per-minute token usage data for PTU sizing.