Token pricing, tracked over time

LLM Price Watch

Every other model comparison answers which model should I pick. This one answers what happened to my AI bill, and why — so the change column leads, and the calculator runs observed repricings against a real workload shape.

No model repriced in the last 30 days.

What a repricing did to your bill

Pick a workload shape and volume; the delta comes from the observed 30-day price change.

No model pricing loaded yet.

Model pricing

USD per million tokens. Sorted by provider, then by input price.

Token pricing for every tracked model, with 30-day price movement.
Model	Input /M	30D	Output /M	30D	Cache read	Cache write	Context	12-mo trend

Prices are published list rates per million tokens. Cache read and cache write are shown where the provider prices them separately; a dash means the provider does not offer that line item rather than that it is free. These rates are not compared against cloud SKUs anywhere on the site: a model billed per token and a Vertex or Bedrock SKU billed per thousand characters are not the same unit, and reconciling them would mean inventing an exchange rate.

Get told when a model reprices

Model pricing moves faster than cloud SKU pricing and usually downwards — which means the cost of not noticing is a bill you could have cut weeks ago.