AI glossary
Terms are defined as this site applies them. Where a term has no standard definition, the entry states the reading used here.
- AGI
- Artificial general intelligence. There is no agreed definition and no agreed test, so this site does not report a distance to it. The alignment page lists the measured quantities that bear on the question. Alignment
- API access
- A release in which a model can be queried over a network but its weights are not distributed. One of three release types recorded for notable models. Capability: how models ship
- Benchmark saturation
- The point at which the best recorded score on a test stops separating leading models. This site uses 90 per cent of the maximum score as the threshold and reports the month it was first passed. Progress
- Binding instrument
- A law, regulation or other instrument that creates legal obligations. Distinguished on this site from guidance, standards and voluntary commitments, which are recorded separately. Policy
- Compute threshold
- A quantity of training compute written into a statute, executive action or official policy document, above which additional obligations apply. Recorded here with the instrument that sets it and the number of models above it. Capability: compute thresholds in law
- Dangerous-capability evaluation
- A test of whether a model gives meaningful help with chemical, biological, cyber or autonomous harm. Counted here only where a card reports results for that model. Alignment: what developers published at each release
- Facility power
- The whole electrical capacity of a datacentre: IT power plus cooling, lighting and power conversion. Datacentres: where they are
- FLOP
- Floating-point operation. Training compute is reported as the total number of operations used in the final training run, on a logarithmic scale, so each axis step is ten times the one below it. Capability
- Frontier model
- A model at or near the highest recorded training compute at its date of publication. The term has no standard definition; where this site plots a frontier it means the running maximum over all models published to that date. Capability
- Frontier safety framework
- A document in which a developer states the capability levels at which it will take specified action. The frameworks are self-written and self-assessed, and share no common scale. Alignment: what developers have promised
- Governance readiness
- A score compiled here from primary instruments, recording the machinery a jurisdiction has for governing AI. It records presence, not quality or enforcement. Methodology
- H100 equivalent
- A unit of compute capacity: the number of Nvidia H100 chips with the same total performance as the chips a site actually holds. Written H100e. Datacentres: every site
- Hourly carbon-free energy
- The share of electricity that was carbon-free in the same hour and on the same grid in which it was used. Google is the only data-centre operator that publishes it. Environment: renewable claims against the grid
- IT power
- The electrical power available to a datacentre's computing equipment, before cooling, power conversion and other overheads. Recorded per site as the power operating now and the power planned at the last stage of the site's timeline. Datacentres: where they are
- Location-based emissions
- Emissions from electricity counted at the average carbon intensity of the grid a data centre drew from, whatever contracts the operator holds. Environment: published footprints
- Market-based emissions
- Emissions from electricity after subtracting renewable energy the operator bought by contract. A company matching all its annual use with renewable purchases can report close to zero while its data centres draw fossil power at some hours. Environment: published footprints
- Notable model
- A model meeting Epoch AI's notability criteria, which is the population used for every model count on this site. Progress
- Open weights
- A release in which the trained parameters are distributed and can be run by a third party. It does not imply an open-source licence, and the licence attached to a given release may restrict use. Capability: the open-weight gap
- Pre-release testing
- Testing of a model by a party other than its developer before public release. Counted here only where a card names a government institute that tested a pre-release version of that model. Alignment: what developers published at each release
- Provenance marker
- The marker beside a figure on this site. It opens the source, publisher, licence, update cadence, retrieval date and stated limitations for that figure.
- Renewable matching
- Buying as much renewable electricity over a year as a company uses, through contracts or certificates. Says nothing about whether the power used in a given hour, on a given grid, was clean. Environment: renewable claims against the grid
- Reported incident
- An entry in the AI Incident Database, which records publicly reported harms involving AI systems. Entries are reports, not adjudicated findings, and the database is not a complete record. Alignment
- Running maximum
- The highest value recorded up to each date. It rises or holds level and never falls, so a line fitted through one cannot slope downward. Capability: the open-weight gap
- Soft law
- Guidance, standards, codes of practice and voluntary commitments that create no legal obligation. Indexed here alongside binding instruments and tagged by force. Policy
- Sui generis database right
- An EU right, separate from copyright, protecting substantial investment in obtaining or presenting a database. It is the reason several third-party policy trackers cannot be bulk-extracted for use here. Sources
- System card
- A document a developer publishes for a model, describing its evaluations, safeguards and limitations. Also called a model card or safety report. Dated here by when it first appeared. Alignment: what developers published at each release
- Systemic risk threshold
- The training-compute level in Article 51(2) of the EU AI Act at which a general-purpose model is presumed to carry systemic risk. It is a rebuttable presumption rather than a fixed boundary. Capability: compute thresholds in law
- Token
- The unit a language model reads and writes: a word, part of a word, or a piece of an image or audio. Each company splits text differently, so a token count from one company is not the same unit as another's. Usage
- Tokens per month
- A disclosed token volume converted to a month by this site, so that per-minute, per-day and per-year figures can be drawn on one axis. A quarterly or annual total becomes a monthly average and is marked as one. Usage: tokens processed
- Working-age population
- The denominator used for the population adoption figures, as defined by the publisher of that dataset. Adoption shares on this site are shares of that population, not of total population. Adoption