StrongREJECT: Results as Published

StrongREJECT measures how much useful information a model gives when asked forbidden prompts that require specific harmful information. The project page groups those prompts into six categories: disinformation and deception, illegal goods and services, hate, harassment and discrimination, non-violent crimes, violence, and sexual content. An automated evaluator scores how useful the reply is, either with a rubric judged by a language model or with a smaller model trained on that rubric. The paper uses that score as the measure of whether a jailbreak produced a genuinely useful answer, checked against human 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.