Independent benchmarks for the price of AI inference, built from verifiable inputs, overseen by an index governing board and published every day.
CICI is the flagship benchmark for the price of inference. CRTI shows what the market actually pays for it. Each is computed from its own inputs and published daily.
The price of AI inference at constant capability, so every move in the index reflects the cost of intelligence itself.
What the market actually pays for inference, in US dollars per million tokens. A view of real costs that complements the benchmark.
Four principles guide every value we publish.
Model prices come from each vendor's own published price page, so anyone can check the inputs behind the number.
As models grow more capable and more affordable, holding capability constant lets the benchmark follow the price of intelligence itself.
The methodology and every change to how the indices are built are approved before they take effect. Each day's values are then calculated and published automatically by that approved process.
Each methodology version is frozen and kept on record. A change to the base methodology is released as a new, named version.
The indices are run under a governance framework modelled on established commodity benchmarks, with clear separation between those who calculate, those who oversee and those who use them.
Enquire about the governing boardIndependent oversight of the methodology, its changes and its use as a benchmark.
Approves the methodology and any change to how the indices are built, and runs the approved process that publishes the indices every day.
The full rulebook is shared with members of the governing board under a confidentiality agreement.
Every published methodology version stays on record and can be retrieved.
Index values and their history are available under licence to institutions, platforms and researchers.
Reference CICI in contracts, hedges and pricing agreements.
Track the cost of inference behind your AI spend, and report it consistently.
A daily record for modelling the economics of AI.
CICI is Carry's flagship benchmark for the price of AI inference. It measures the price of inference at constant capability, so every move in the index reflects a change in the cost of intelligence itself. It is designed as a reference for hedging and settlement.
CRTI measures what the market actually pays for AI inference, in US dollars per million tokens, using realised prices and observed usage. It is computed from its own inputs, independently of CICI, and is designed for market monitoring and research.
Model prices are taken from each vendor's own published price page, so anyone can check the inputs behind the published values.
Both indices are published daily.
An index governing board provides independent oversight of the methodology, its changes and its use as a benchmark. The methodology, and any change to how the indices are built, is approved before it takes effect. Each day's values are then calculated and published automatically by that approved process. Each methodology version is frozen and kept on record.
The full rulebook is shared with members of the governing board under a confidentiality agreement.
Index values and their history are available under licence for benchmarking and settlement, risk and reporting, and research. Contact [email protected] to discuss a licence.
Carry Exchange is where participants trade the AI value chain, from compute to the commodities behind it.