AI Vision Index, private tests of how well 36 AI models see
Source
aivisionindex.com
Author
LLD
Date
Why it matters
Public vision benchmarks can be trained on. This index uses held-out tests and reports price and speed alongside scores, so you can pick a vision model on uncontaminated results.
Key takeaways · AI-distilled
The author built private tests because public vision benchmarks publish their datasets, which models may have trained on; every question has one specific answer, and the tests will be replaced after they are used.
In the author's results, GPT-6 Astra leads at 83.7, with GPT-6.1 Sol close behind at 81.3 for a noticeably lower price; Gemini 4 Argon and Haiku 5.5 are described as very cheap and quite good.
Some tasks easy for people stay hard for models: counting more than 20 objects, and spotting obviously AI-generated images, where the author says most models are barely better than a coin flip, Argon excepted.
Terms in this piece · Glossary
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
