🧭 Model Release Map · 2026 → 2017Open the full page ↗
Interactive map · Jun 2017 → Nov 2026

Model Release Map

Who shipped which foundation, image, video, speech and world model, and when. Each lane is a lab, each line joins a model to its successor, and dashed outlines mark open-weight releases, so you can see how fast the labs move and how long open models take to catch up.

How to read this, and what it does not cover

  • Dates are each source's announcement or first public availability. Month-only dates are kept as months. Research models that were announced but never released (Transformer, Meena, Make-A-Video, Movie Gen, Voicebox) are shown faded. Before 2022 the date is the paper's or the lab's own announcement.
  • Foundation models cover 13 labs, are filled in through 17 Nov 2026, and reuse the entries on the Frontier Lab Updates page. Image, video, speech and world models add specialist labs, but the list is a selection, not every release, and speech, video and world models are thinner than the rest. World models (interactive or predictive simulators of an environment, for games, robots and driving) are tracked from Wayve's GAIA-1 in June 2023.
  • Sources: entries marked “secondary source” rest on press or reference pages, not the lab's own announcement. Treat them as less certain.
  • Open means the weights are public, not that the licence is permissive; “not recorded” means this dataset has not checked the licence.
  • Open vs closed compares the first release at each capability within this dataset. An earlier model outside it would change the lag.
  • OpenAI shut down the Sora app and web experience on 2026-04-26 and scheduled the API removal for 2026-09-24, so Sora is drawn struck through.
  • Leaderboard scores are not copied here because they go stale quickly; check the labs' own pages for benchmarks.