CorpFin: Results as Published

CorpFin (CorpFin v2) tests whether a model can answer questions about long-context credit agreements, contracts that are often over 200 pages long. Questions cover extracting terms and numbers, summarizing, numeric reasoning, cross-referencing earlier definitions, judging terms against market standards and understanding industry jargon. The private test set has 858 questions from 43 documents, written by Vals AI with financial analysts, legal professionals and academics. Answers are graded by an Anthropic model as the LLM judge at temperature 0, under three context setups (Exact Pages, Shared Max Context and Max Fitting Context). The printed metric is accuracy, as a percentage, and the benchmark is now archived because performance has saturated.

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.