Financial QA Benchmark
Which LLM should power your document agent? Each model answers the same 100 hard questions about 40 annual reports (Form 10-K), with PageIndex doing the retrieval; only the model changes. See the benchmark's GitHub repo for details.
Updated
Accuracy vs. Cost per Run
Share of the 100 questions answered fully right, against the cost of one run through all of them
Pareto frontier
Scores move by a few points from run to run, so treat a gap of one or two points as a tie.