VLMs are Blind: Results as Published

VLMs are Blind, also called BlindTest, checks low-level vision that a person can settle at a glance. The project page lists seven tasks: counting how many times two lines meet, judging whether two circles touch or overlap, naming the circled letter in a word, counting overlapping shapes, counting nested squares, counting cells in a grid, and following a path on a subway-style map. Each item has a fixed set of possible answers. The reported result is accuracy on a task, and the page also prints a mean across the seven tasks.

Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.