New Free 'Radar' Lets Anyone Check If AI Claims Are Real
One website now tracks 12,916 AI test scores — so hype is easier to spot.
A group of researchers has launched Benchmark Radar — a free, constantly updated website and search engine that collects the standardised tests used to grade AI systems. Benchmarks are simply exams for AI: a set of questions or tasks a model must complete, scored out of 100. Companies regularly announce that their AI is "the best," but those claims rest on exams that are hard to find, hard to compare, and sometimes quietly retired when a model does badly. Benchmark Radar pulls fresh test papers, datasets and code from 37 sources every single day.
The catalogue is already substantial: 1,283 source records from four separate benchmark collections, plus 12,916 individual scores logged across 790 entries. It includes a leaderboard of top performers, a chart showing which AI delivers the most useful work for the least computing power, and trend views that reveal which tests are becoming popular — or quietly abandoned. One important feature flags "saturated" benchmarks: tests where nearly every major AI now scores close to perfect, so the test can no longer tell models apart. Every entry keeps its original source and citation, so you can trace a score back to the evidence behind it.
Why does that matter outside research labs? If you are deciding which AI assistant to buy for your team, or reading a press release claiming a world-record result, this gives you a way to check whether the test behind that headline is still meaningful. A company bragging about acing a 2023 exam that everyone has since maxed out is not the same as one topping a hard, current test. Journalists, policy staff, procurement teams and curious buyers all get a shortcut that previously only insiders had.
The catch: Benchmark Radar is built for researchers, not shoppers. It has no product reviews, pricing or plain-English buying advice, and a great exam score still does not guarantee an AI will write your emails or handle your customers well. It is also a research project, so coverage can be uneven. Treat it as a hype detector, not a shopping guide.
- Benchmark Radar is a free, constantly updated website that gathers the standardised tests used to grade AI, pulling from 37 sources every day.
- It already holds 1,283 records and 12,916 individual scores, and flags tests that every AI now aces — meaning they no longer prove anything.
- Every score links back to its original source, so you can check the evidence behind a company's 'best AI ever' claim.
Why It Matters
Could help you tell a genuinely better AI from good marketing before you pay for it.