About this changelog
This changelog is a dated, reverse-chronological record of notable changes to Next AI Drop: feature launches, methodology revisions, and data corrections such as a shifted expected window or a resolved codename. Newest entries sit at the top. Every entry is factual and links to the part of the product it touched.
It also works as a freshness and trust signal. The product's edge is hourly cached intel, so a visible, honest record of what moved — and when a forecast was wrong — matters as much as the forecasts themselves. When an expected date slips or a codename maps to a real model, that correction is logged here rather than quietly overwritten. See the sources and correction policy for how corrections are decided and cited.
Keep two logs distinct in your head. This page tracks the product: pages, features, and how the score is computed. The model releases themselves — what shipped, what is imminent, what is still rumoured — live on the live forecast board and refresh hourly. Nothing here restates that feed.
2026-05-29 — Launch: forward forecast timeline
Launched the core product: a release timeline that extends past today into forecast windows, so the page answers "what is next" instead of only "what shipped". Most trackers stop at the present; this one continues into dated forecast windows for models that have not been released yet.
Each forecast carries a stage — imminent, expected, or rumoured — and a target window rather than a single hard date. Open the live forecast board to see the current ranking and how far ahead each window reaches.
2026-05-29 — Launch: Drop Readiness score
Introduced Drop Readiness, a single 0–100 score that ranks how close a model is to shipping. The formula is fixed and public: DR = .45 odds + .25 intel + .20 deadline + .10 volume. One number per release means you can sort and plan against it directly.
The weights are deliberately stable so the score reads the same way week to week; any future change to them will be logged as its own dated entry here. For the full computation, worked example, and what each input means, read how Drop Readiness works.
2026-05-29 — Launch: cross-venue odds (Polymarket + Kalshi)
Blended forecast-market odds from Polymarket and Kalshi into a single odds input and surfaced the cross-venue spread between the two. Two venues priced side by side is a sharper signal than either alone, and the gap between them is informative in its own right.
Markets are signal, not stakes — the odds feed the forecast, never a wager. Polymarket and Kalshi are the sources we build on; we add the second venue, the spread, and the rest of the score on top. The odds component is the .45 term in the Drop Readiness formula.
2026-05-29 — Launch: hourly intel
Launched an intel feed that refreshes hourly, with every item tagged Confirmed, Rumor, or Market and linked to its source. The tag tells you how much weight an item carries; the link lets you check it yourself. Hourly cadence is the differentiator against multi-day editorial trackers.
Intel is drawn from official changelogs, public posts, public model-catalog listings, and market moves — each item source-linked so you can verify the claim, not just take ours. See sources and correction policy for what qualifies and how a tag is assigned.
2026-05-29 — Launch: codename watch
Added a dedicated codename watch that maps internal pre-release names to the public models they likely become. Codenames such as ⟨MYTHOS⟩ or ⟨EMBER-ALPHA⟩ surface in public posts long before a product name does, and connecting the two early is a real head start for builders.
Each mapping carries a confidence level and resolves to a confirmed entry once the real model ships — and when a mapping turns out wrong, that gets a correction entry here. Browse current decodings on the codename watch. Examples above are illustrative, not live mappings.
2026-05-29 — Launch: documentation and trust pages
Published the documentation and trust set: methodology, glossary, sources, FAQ, and the legal pages (privacy, terms, cookies, disclaimer). These exist so every number on the product is explainable and every claim is checkable, which is the point of a forecast you are meant to plan around.
Start with how Drop Readiness works for the full computation, or the frequently asked questions for short answers to the common ones — including why a forecast tool deliberately is not a betting product.
Next AI Drop is a forecasting tool for AI builders. Prediction-market odds from Polymarket and Kalshi are used only as a signal of when a model is likely to ship — never as betting, gambling, or financial or investment advice. Entries below describe product changes, never how anything was detected.
How to follow updates
Future changes — new features, methodology revisions, and data corrections — will be logged on this page and pushed to the RSS feed, so you can track product changes the same way you track releases. Entries stay factual: what changed, on what date, and which part of the product it affects.
Corrections are first-class here. When an expected window shifts or a codename resolves, it becomes a dated entry rather than a silent edit — that is the honesty standard a forward-looking forecast has to hold. The visible "Last updated" line at the top of every page and the dateModified in its structured data move together, so freshness is never overstated.