OpenAI cut Luna 80% — Azure kept the old price, so Luna costs 5× more on Azure than direct

OpenAI cut GPT-5.6 Luna by 80% on July 30. We covered it. What we did not check — and should have, because it is the number that actually lands on a bill — is whether the clouds passed it on.

Azure did not. Read from the live catalog this morning:

Luna, standard tier. OpenAI direct $0.20 in / $1.20 out per 1M tokens. Azure $1.00 in / $6.00 out. That is exactly 5× on both sides of the meter.

Luna, the long-context tier above 272K. OpenAI direct $0.40 / $1.80. Azure $2.00 / $9.00. Again 5×.

Terra. OpenAI direct $2.00 / $12.00. Azure $2.50 / $15.00 — a 25% premium, which is ordinary cloud markup rather than a refused discount.

Sol. $5.00 / $30.00 on both. Identical, because Sol never got a cut to withhold.

Azure Cognitive Services carries the same figures as Azure proper, so switching between the two endpoints changes nothing.

The practical consequence is narrow and expensive. Anyone who read the July 30 headline, budgeted a Luna workload against $0.20/$1.20, and deploys on Azure for procurement or compliance reasons is out by a factor of five. On a million-token-per-day pipeline that is the difference between roughly $0.20 and $1.00 a day of input — trivial — and between $1.20 and $6.00 of output, which stops being trivial the moment the workload is real.

We are stating what the catalog says today, not why. Microsoft has published no explanation for holding the pre-cut price, and a cloud reseller is under no obligation to track a model provider's list price. But "OpenAI cut Luna 80%" and "Luna got cheaper for you" are different claims, and only the first one is true everywhere.