About AI Radar
A personal, non-commercial project by Benjamin Nieuwenhuijzen.
AI Radar is a live dashboard that collects the latest news and updates from the leading AI companies on one page. Every post links straight to the original article. Next to the news, Model History records which AI models those companies released, when, and what changed, with the sources behind every fact.
What the dashboard shows
At the top, the Today in AI carousel shows a handful of top stories, picked automatically from the latest posts and labelled Auto-selected. Next to it, the Latest column lists the five newest posts. A live status line shows when the data was last updated, counts down to the next scheduled refresh and gives the number of sources.
Each company has its own card, from OpenAI and Anthropic to Mistral and DeepSeek. The Across AI card adds industry news that is not tied to a single company. You can switch between two views, By company and Timeline, narrow the posts down with the category chips or search, and save posts for later. The Motion toggle turns off all movement on the page; this also happens automatically when your system asks for reduced motion.
Where the posts come from
AI Radar reads publicly available RSS and Atom feeds. Each card states its source: an official feed is maintained by the company itself, a community feed is maintained by others and may lag behind. A scheduled job fetches all feeds every two hours and rebuilds the page data. Nothing is rewritten or shortened by hand; posts appear as their source published them.
Model History
Model History is AI Radar's own content: a history of the models of the same companies, with a page per company, a page and a timeline per model, and the relations between models (which one succeeded, revised or was derived from which). Dates are shown exactly as precisely as the sources give them: if a source names only the month, the page shows only the month. Every fact links to its sources, and each source shows whether the original is still online or only available as an archived copy.
The records are compiled from the organisations' own announcements, model cards and documentation. The summaries and change notes are drafted with the help of an AI assistant from those sources and checked against them. Newly released models are added by an AI agent (see below). It is not a benchmark or a ranking: there are no scores, and claims made by a company are attributed to it.
What is automated, and therefore fallible
Several things on the site are produced by software rather than a person, and each one can be wrong:
- Categories (Model releases, Products, Research and so on) are assigned from a post's title and summary. A mislabelled post is an estimate gone wrong, not an editorial judgement.
- Card order follows recent news activity.
- The "Today in AI" stories at the top are picked by a simple rule: recent model releases first, at most one per company. No AI model writes or rewrites them; each slide shows the post's own title and summary.
- Source status in Model History ("Active", "Archived copy", "Unavailable") comes partly from a daily automatic check of each source address, run on GitHub's servers. A site that blocks the check or is briefly down can be marked wrongly for a while.
- Related AI Radar coverage on a model page is linked automatically when a post's title or summary contains one of the model's names. A link can be missing or, now and then, point to the wrong model.
- New models in Model History: when AI Radar posts name a model that has no page yet, an AI agent looks up the organisation's own sources once a day, writes the record and its summary, and publishes it after automatic checks of the data. No person reads these entries before they appear, so a new entry can contain a mistake until it is corrected.
For anything that matters, click through and read the original article.
Who makes it
AI Radar is made and maintained by Benjamin Nieuwenhuijzen. It is not affiliated with, or endorsed by, any of the companies it follows. All rights to the posts remain with their original sources, as set out in the Disclaimer. The dashboard needs no account and collects no personal data about you; the details are in the Privacy Statement.
Suggestions for new sources, corrections and other feedback are welcome via the contact page. Prefer your own reader? There is an RSS feed with everything combined.