Budget time-limited AI discounts twice to avoid the surprise invoice

The discount has an end date

Your AI budget can include a rate that expires. On August 21, 2026, OpenAI lowered GPT-5.6 Sol's standard API input rate to $4 per million tokens and output rate to $20 per million tokens, down from $5 and $30. OpenAI treated the lower rates as a promotion available at least through November 21, 2026. The cut was a 20% input-price and 33.3% output-price reduction.

For a small team, the spread is the number to track: the gap between the promotional bill and the bill after the promotion ends.

Track the spread, not the list price

A workload of 1,000 calls, each using 4,000 input tokens and 800 output tokens, would cost $32 with the new GPT-5.6 Sol rates instead of $44, while Claude Opus 5 would cost $40. Anthropic's Claude Opus 5 was priced at $5 per million input tokens and $25 per million output tokens. The promotional Sol rates were $1 per million input tokens and $5 per million output tokens below Claude Opus 5. The budget needs the live-promotion cost and the post-expiry cost side by side.

Build the two-rate budget before you commit

Use the same workload for both rates. Do not model the promotional rate with a small demo workload and the post-expiry rate with the full production workload.

• Separate the promotional rate from the post-expiry rate. Write both rates in the budget sheet, with the expiry date next to the promotional rate. Finish the step when someone can read the sheet and see the rates separately.
• Model the workload at both rates. Use the actual call count, input tokens, and output tokens from the last billing period. A cost at the promotional rate and a cost at the post-expiry rate should appear in the sheet.
• Log the expiry in the ops calendar. Put the date in the shared calendar, assign an owner, and set a review date before the discount ends. The task should be visible to the owner without a question.
• Set the decision date before the discount ends. Decide whether to renew, switch models, reduce volume, or accept the higher rate. A written recommendation and an updated budget line should close the step.

If the vendor has not published the post-expiry rate, use the pre-promotion rates as the conservative planning number until you get a written answer from the vendor.

The discount reached eligible token-metered activity in Codex and ChatGPT Work, while Plus, Pro, and Business subscription fees, included usage, and five-hour/weekly caps stayed the same.

Keep the sheet small. A row for the promotional rate, a row for the post-expiry rate, and a row for the difference. Add the model name, the token mix, the call count, and the expiry date. If the team changes the prompt or adds a new workflow, update the token mix before the next billing cycle.

Make the expiry a review, not a surprise

The post-expiry rate shows up in the invoice. The calendar entry forces the team to look at usage, the rate, and the decision before the discount disappears.

At the review, ask the right questions. Is the workload still worth the post-expiry cost? Is there a cheaper model that meets the quality bar? Can the team reduce calls, shorten prompts, or batch work? Change the budget line, not just the spreadsheet.

Produce a written note: stay, switch, reduce, or renegotiate. Keep the note short, dated, and final. If it says switch, record the new rate and the migration owner.