We-Math: Results as Published

We-Math tests visual math reasoning beyond the final answer. It collects 6.5K diagram problems spanning 67 knowledge concepts in five layers, and it splits each multi-step problem into the one-step sub-problems those concepts require. The model is scored on both the steps and the full problem, then placed in one of four outcomes: insufficient knowledge, inadequate generalization, complete mastery, or rote memorization. The project page also defines a strict and a loose version of that judgment, and it recommends the strict score on its 1,740-example testmini set.

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.