OpenAI halved GPT-6 token prices on September 22. A list-price cut rarely flows straight through to your AI bill. Here is where the savings go missing and what to do about it.
Category: AI Cost Management
Azure OpenAI’s Flex Tier Cuts Token Costs 50%. It Has No Safety Net.
Azure OpenAI's new Flex tier bills tokens at half the Standard rate for latency-tolerant work. The discount is real; keeping it depends on a routing decision no API makes for you.
Cloud FinOps Was Built on the Server-Hour. Token Economics Just Broke the Model.
The FinOps Foundation renamed its flagship event after AI tokens. The atomic unit of cost changed, and the tooling finance teams built to meter the last one cannot see the new one.
Writing to the Prompt Cache Used to Be Free. On GPT-5.6, It Costs 1.25x Input.
Azure OpenAI and OpenAI now bill prompt cache writes at 1.25x input on GPT-5.6. Here is the break-even math and the workloads the change quietly taxes.
Your AI Dashboard Measures Tokens. Your CFO Wants Cost Per Outcome.
The FinOps Foundation's June 2026 token framework lists four AI unit metrics. None is cost per token. Why cost per outcome is the number that survives agents.
The Better Your Team Optimizes AI Spend, the Less Your Forecast Is Worth
Model prices drop every 6 to 9 months and a single routing or compression change can cut a line item 85% to 99%. Here is why AI cost forecasting breaks the cloud playbook, and what to forecast...