When a Price Cut Made a Model 14x Bigger: What the GPT-5.6 Discount Data Actually Shows
- David Borish
- 11 minutes ago
- 4 min read
What the discount did to volume
OpenRouter measured three windows: a pre-period from July 8 to July 26, the discount program from July 27 to August 14, and a short post-period from August 15 to August 20. Against the pre-period daily average, Luna tokens climbed 13.8x during the program and Terra tokens 5.6x. The discount stacked over time. OpenAI cut its own list prices on July 30, dropping Luna 80% and Terra 20%, and those cuts landed on top of an existing 50% OpenRouter discount. From July 30 onward the effective reductions reached roughly 90% for Luna and 60% for Terra.
Sol serves as the useful contrast. It stayed at list price through August 16 and moved only 1.11x over the same stretch. Other OpenAI models fell slightly. Models outside the OpenAI family rose gently. The only lines that bent sharply upward were the two that got cheaper, which is what makes the price mechanism legible rather than speculative.
Where the new tokens came from
A volume spike alone does not prove new demand. Some of it could be existing OpenAI users shifting from one model to another, which would show growth for Terra and Luna without expanding OpenAI's overall footprint. OpenRouter's share data addresses this directly. Terra and Luna went from 0.7% of all OpenRouter tokens before the program to 7.8% during it, a gain of 7.1 percentage points. Competitors outside OpenAI gave up 5.3 points and other OpenAI models gave up 1.9 points. Roughly three quarters of the gain came from outside OpenAI rather than from cannibalizing sibling models.
Across the whole OpenAI family, token share grew from 7.1% to 12.4% and crested above 15% on some days. OpenAI's total token volume on the platform nearly doubled and held at that higher level after the program ended. One author's tokens moved the other way. Anthropic's volume declined over this timeframe, according to OpenRouter's by-author chart, though the post-period is only six days and too short to read as a durable trend.
Whether the users stayed
The more interesting question for anyone modeling AI demand is retention. A discount that empties out the moment it expires tells you about promotion sensitivity rather than any lasting shift in behavior. OpenRouter tracked the more than 100,000 customers who used Terra or Luna during the program. About 32% kept some usage in the days after the discounts ended, and 18% ran at or above their program pace. That is a customer count, not weighted by tokens.
Weighting changes the picture. Measured by daily token volume, the post-program period ran at 1.38x the discount period's average, which means the accounts that stayed were far larger than the median program user. The heavy users were the ones who stuck. OpenRouter is candid that six post-program days against a 19-day program is thin, and the retention story could shift as more data accumulates.
Reading the Jevons framing carefully
OpenRouter attached the Jevons paradox to these results, and the fit is reasonable at the level of the individual model. Cheaper Luna tokens produced more than proportionally more Luna usage, and revenue behavior reported elsewhere is consistent with that. Analysts at TD Cowen, cited in secondary coverage, estimated that OpenAI's Luna revenue rose roughly 34% versus the week before the cut despite the price falling to about a tenth of its prior level, with Terra revenue up around 45%. Revenue climbing after an 80% price cut is the kind of result that makes the Jevons label stick.
The framing gets shakier when it stretches to the whole system. The classic paradox describes total consumption of a resource rising as efficiency improves across an entire economy. What OpenRouter measured is one gateway over a few weeks, with a large share of the gain coming from users switching away from competitors rather than from brand-new work that did not exist before. Displacement and induced demand are both present in the data, and OpenRouter's own numbers let you separate them: 7.1 points of share gained, 5.3 of which came from rivals. Calling the whole thing Jevons flattens that distinction.
Why the gateway matters more than its size
There is a second reason to read these charts with care, and it has nothing to do with the token math. OpenRouter and Vercel's AI Gateway handle a small fraction of OpenAI's total traffic, but they are among the few public windows the industry uses to estimate model market share. SemiAnalysis raised this point when Sol received its own 50% discount on August 17, noting that a temporary promotion can double a model's visible volume on exactly the dashboards investors watch, which risks reading a discount response as a genuine competitive win.
That concern applies to the Terra and Luna data too. The usage surge is real routed traffic, and the retention among heavy users is a meaningful signal. But a discount that moves the most-watched public gauge is going to shape narratives about who is winning, and the gauge does not show the whole market. The context worth keeping in view is competitive: Chinese-origin and open-weight models have held above 30% of U.S. OpenRouter token share every week since February 2026, cheap enough that routing decisions increasingly turn on price. OpenAI's discounts landed in that environment, which is part of why they worked and part of why they are hard to read as a clean market-share verdict.
What this shows and what to watch
The clearest takeaway is behavioral: token traffic on these gateways moves fast and hard when the meter gets cheaper, and at least among larger accounts, some of that movement persists after the promotion ends. The demand for capable model tokens is elastic in a way that price cuts can activate quickly. That is a useful thing for anyone forecasting AI spend to know, and it is well supported by the volume, share, and retention data OpenRouter published.
The claims worth holding loosely are the broad ones. The 13.8x figure is Luna alone, not Terra and Luna together. The retention window is six days. The revenue estimates come from third-party analysts rather than OpenAI's own reporting. And the gateway that produced these charts is a narrow, heavily watched proxy that a discount can distort. The next number to wait for is a settled Sol multiplier once its discount window closes, alongside a longer post-period that shows whether the retained users of Terra and Luna keep their pace or drift back once the novelty and the savings both fade.
