By Q1 2027, OpenAI cuts the per-token API price of its flagship model by at least 30 percent

Prediction Statement

By March 31, 2027, OpenAI will reduce the published per-token API price of its current flagship model — the premium, top-of-menu general model, not a new cheaper tier — by at least 30 percent relative to its price on June 30, 2026, on at least one of input or output tokens. A new budget SKU launched alongside an unchanged flagship price does not satisfy this prediction; the cut must land on the premium model that enterprise default traffic actually runs on today. A temporary promotional discount does not count either — the reduction must be a standing change to published list pricing.

I am putting tier-2 confidence at 68 percent.

Reasoning and Analysis

The prediction follows directly from the demand-side repricing now underway. As I argued in The Efficiency Turn, the largest AI customers have begun moving from tokenmaxxing — routing everything to the best model — toward governed efficiency, downshifting routine traffic to budget-tier models a fraction of the price. Reporting in late June 2026 named the trend explicitly, with Uber said to have burned its annual AI budget in four months and the CEO of Lindy moving all of his company traffic off Claude to a cheaper model.

When customers can credibly migrate flagship traffic to a model an order of magnitude cheaper without users noticing, the flagship price loses its anchor. The budget tiers — DeepSeek V4 Flash near fifteen cents per million input tokens, plus the cheaper SKUs from Microsoft MAI, Google, and OpenAI itself — set a reference point that makes the premium model look expensive on the broad middle of tasks. Reporting has already surfaced that OpenAI is weighing deep price cuts as token economics hit an inflection point. The competitive logic is hard to escape: defend the flagship by making it cheaper, or watch the volume drain to the budget tier and to rivals.

The confirming mechanism is OpenAI's confidential IPO filing. A company heading to public markets wants to show durable, defensible volume rather than a revenue line exposed to customer-side optimization. Cutting flagship pricing to retain and grow volume is a rational pre-IPO move, even at the cost of revenue per token.

What Would Falsify This

The prediction fails if, on March 31, 2027, the published per-token list price of OpenAI's flagship model is unchanged or higher than its June 30, 2026 level on both input and output, with any price relief delivered only through a separate cheaper model rather than a cut to the flagship itself. It also fails if the only reductions are temporary promotions that revert.

The main risk to the prediction is a capability step-change: if OpenAI ships a new flagship that opens a large quality gap over every budget tier, it regains pricing power and has no reason to cut. A second risk is that the company segments instead of cutting — holding the flagship price while pushing efficiency-seeking customers onto cheaper SKUs, which would let it sidestep a headline flagship reduction entirely. Both are plausible, which is why this sits at tier-2 rather than tier-1 confidence.