Glossary

This glossary defines the forecasting and AI-release terms used across Next AI Drop, so a single line is enough to understand a score, a stage, or a tag. Terms are alphabetized below with anchor links, and each is written as a standalone definition you can quote on its own.

Benchmark frontier

The current best score on a benchmark together with the challenger expected to take the lead next, including that challenger's projected timing. It answers two questions at once: who holds the top score today, and which model is likely to overtake it and when.

Codename

An internal or pre-release name a provider uses before a model's public name is set, often surfacing in public model-catalog listings or executive posts. We map each codename to the public model it most likely becomes. Codenames render bracketed and uppercase in mono — for example ⟨MYTHOS⟩ — across the site and the codename watch; see codenames decoded for how a name becomes a launch.

Context window

The maximum amount of text a model can consider at once, measured in tokens — for example 1M. A larger window lets a model reason over more input in a single pass without losing earlier context.

Cross-venue

Spanning both prediction-market venues we source from, Polymarket and Kalshi, blended into one signal rather than read in isolation. Reading a release across venues smooths single-market noise and exposes disagreement; see the related odds spread.

Drop Readiness

A 0-100 score estimating how close a model is to shipping, computed as DR = .45 odds + .25 intel + .20 deadline + .10 volume. It compresses market odds, intel recency, deadline proximity, and market volume into one number you can plan around; see how Drop Readiness is calculated for the full computation.

Forecast window

The date range within which a model is most likely to ship, derived from expected release dates and market resolution dates. It frames a launch as a window rather than a single guessed day.

Imminent, expected, rumoured

The three release stages a tracked model moves through. Imminent means high readiness and near-term timing; expected means a credible window backed by solid signal; rumoured means early or single-source, still unconfirmed. Stages follow from the score described in the methodology.

Intel tag (Confirmed / Rumor / Market)

The provenance label on an intel item. Confirmed marks official or otherwise verifiable information, Rumor marks credible but unverified reports, and Market marks a read derived from a market move. Every item is source-linked so the tag can be checked.

License (open / proprietary)

Whether a model's weights are openly available (open) or access is restricted to a provider's API or product (proprietary). The distinction sets what you can self-host versus only call remotely; license appears on each entry in the model database.

Odds spread (cross-venue spread)

The gap between Polymarket and Kalshi implied odds on the same release. A widening spread is itself a signal — venues disagreeing on timing often precedes news — and feeds the odds input described in the methodology.

Prediction market

A market whose prices reflect the crowd's implied probability of an event: a price of 80c implies roughly an 80% chance it resolves yes. We read Polymarket and Kalshi prices as forecasting signal only — markets are signal, not stakes, not advice to trade.

T-Nd

Days-to-expected-release notation: T-19d means a model is expected in roughly nineteen days. A smaller N raises the deadline component of Drop Readiness, pulling the score toward imminent.