OpenAI reduced announced GPT-5.6 Sol pricing on August 21, 2026, to $4 per million input tokens and $20 per million output tokens. The release record describes this as promotional pricing available at least through November 21. For operators, the decision is whether the whole workload becomes more economical under the applicable terms.
Treat the discount as a dated commercial condition
The changelog is the source for the historical rate and promotion window. Use the current pricing reference when preparing a deployment budget. “At least through” a date should not be converted into a permanent rate or an invented fixed expiration.
A product with recurring revenue needs more than a discounted trial estimate. Calculate the margin under the currently applicable price and under a documented non-promotional scenario if one is available. If the later rate is not established, record that uncertainty rather than treating the promotional number as a long-term guarantee.
Measure the task rather than the final answer
Include all model calls required to produce an accepted result: retries, follow-up reasoning, verification, and human correction. Account for any tool charges and processing-tier differences. This is especially important for agents, where the visible final response can be only a small part of the work.
Compare the same task set and acceptance standard before and after a routing change. A higher-capability model can be economical if it avoids repeated attempts, but that needs evidence. A lower token price by itself does not establish the result.
Use the promotion to learn, then choose deliberately
The discount can make a bounded evaluation less expensive. Start with work whose outputs can be reviewed and whose cost is observable. Keep the current route for tasks already meeting their quality and budget targets until the candidate demonstrates a practical improvement.
Set an owner and a review date for the promotional assumption. Pricing changes should lead to explicit configuration and budget decisions, rather than an unnoticed change in a spreadsheet that later becomes a production margin problem.