How fast have LLM prices actually fallen?

For comparable capability, flagship token prices have dropped roughly an order of magnitude over the past couple of years — GPT-4-class input pricing went from tens of dollars per million tokens to low single digits as GPT-4o and competitors arrived. The tool fits this to an exponential curve and reports the effective halving period.

Why fit an exponential instead of a straight line?

Technology cost curves are multiplicative, not additive — prices tend to fall by a roughly constant percentage each period, which is exponential decay. A straight line would project negative prices; an exponential asymptotes toward zero, matching how compute and model efficiency improvements compound.

Can I trust the forward projection?

Treat it as a planning scenario, not a guarantee. The trend captures historical momentum, but prices can plateau (frontier capability gets harder) or drop faster (new architectures, competition). Use it to bound a budget range, not to commit to an exact future invoice.

Can I enter my own data?

Yes. Paste your own (YYYY-MM, price) points and the same regression runs on them, which is useful for tracking a specific model family or your negotiated enterprise rate over time.

No. The regression runs entirely in your browser. Nothing you enter leaves your machine.

LLM Price Drop History Calculator

LLM price drop tracker

Token prices for frontier models have fallen dramatically and consistently. If you are budgeting a product 12-24 months out, assuming today’s price is almost certainly too pessimistic. This tool fits the historical decline to an exponential curve, tells you the effective price-halving period, and projects where costs land at your chosen horizon.

How it works

The tool takes a series of (date, price-per-million-tokens) points — either the built-in flagship track or your own pasted data — and fits an exponential decay of the form price(t) = a · e^(−k·t) by linear regression on the log of price against months elapsed. From the decay constant k it derives:

the halving period ln(2) / k (how long until price halves),
the annual decline rate (1 − e^(−12k)), and
the projected price at your chosen horizon.

Fitting in log space is the right move because price declines compound multiplicatively, so a straight line in log space corresponds to a constant percentage drop per month.

Tips and notes

A short halving period (under a year) means you should budget conservatively and re-price your margins regularly — your cost base is moving under you in your favour.
Frontier capability tends to plateau in price even as last-generation capability gets cheap; the steepest declines are usually for “good enough” models, not the absolute newest one.
Use your own enterprise-negotiated rates as the input series to see whether your contract is keeping pace with the public market trend.
Projections are scenarios. Pair the optimistic exponential with a flat-price worst case to bound your budget rather than betting on a single number.