GDP.pdf: Results as Published

GDP.pdf tests whether a multimodal model can answer the questions working professionals ask about a specific PDF, such as a lease, a datasheet or a clinical guideline, with the answer grounded in the document. It has 100 items written by professionals in ten fields; an item was kept only if at least two frontier models failed it in a meaningful way. Each item is tagged against eleven capabilities (text extraction and grounding, table and chart comprehension, cross-referencing, spatial reasoning, abstaining on unsupported queries) and scored against a rubric of atomic criteria. Publishers report the strict task-level pass rate ("all pass rate"), the share of items answered well enough to pass every criterion.

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.