FACTS Grounding: Results as Published

FACTS Grounding tests whether a long answer stays inside a document supplied in the prompt and still fulfills the user's request. Each example pairs a human-written request with a document of at most 32k tokens. The benchmark has 860 public examples and 859 private examples. Judge models first disqualify a response that does not answer the request, then mark it accurate only if it is fully grounded in the document, and the factuality score aggregates those judgments.

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.