What happened when GPT 5.6 Terra and Luna were heavily discounted on OpenRouter?
Token usage exploded by 13.8x
Jevons Paradox = as technology makes the use of a resource more efficient, total consumption of that resource increases rather than decreases 🧵
Some of the additional tokens were from older OpenAI models.
But the bulk of the increase were net new tokens + stealing share from competitors.
Importantly, many of the users with Terra/Luna usage during the discount window kept using the models after the discounts were retired.
Overall tokens post-discount were actually higher on a daily average basis than during the window (mostly thanks to a few whales)
More insights on the impact of these discounts in the full piece: https://t.co/9mjQz0zBgu
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
Why it matters
inferenceRunning a trained model to get answers — the phase where AI is actually used, as opposed to trained.Full definition → demand is steeply price elastic and the usage sticks: a temporary cut moved Luna volume 13.8x and left daily averages higher after prices returned, which matters for anyone forecasting tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → spend or capacity.
Read the full thread on X ↗
