Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Agentic benchmark
Refusal of harmful cyber requests (JailbreakBench, StrongREJECT, AgentHarm). Every model the catalog carries a published Cyber Misuse Refusal value for, ranked by that value.
Average refusal rate on the cyber-related prompts of JailbreakBench, StrongREJECT and AgentHarm, as measured by Mistral. Higher means the model refuses more malicious cyber requests. 4 models in the ModelChoice catalog have a published value.
Each point is a model with a published Cyber Misuse Refusal value and complete token prices, on the 75% input / 25% output blend. Frontier models are spaced evenly along the cost axis.
4 models with a published Cyber Misuse Refusal value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.
| Rank | Model | Provider | Average refusal rate |
|---|---|---|---|
| 1 | Mistral | 96.6 | |
| 2 | Moonshot AI | 94.0 | |
| 3 | Z.AI | 87.0 |
What the benchmark measures and how its value is published, from the source’s own record.
Cyber Misuse Refusal (Refusal of harmful cyber requests (JailbreakBench, StrongREJECT, AgentHarm)) is an agentic benchmark published by Mistral Large 4 is now open weight. Average refusal rate on the cyber-related prompts of JailbreakBench, StrongREJECT and AgentHarm, as measured by Mistral. Higher means the model refuses more malicious cyber requests. Its evaluation set is described by the source as cyber-related harmful prompts from three safety benchmarks. The source rates its difficulty as cyber misuse safety.
Each model's result is published by Mistral Large 4 is now open weight as average refusal rate. ModelChoice prints that value to 1 decimal place, exactly as published, and never converts or rescales it. This page ranks the 4 models that carry a published value, highest value first; a model the source never scored is left out rather than ranked at zero. The source does not state whether a higher value is the better result, so neither does this page.
| 4 | 84.0 |
Evidence key: Observed
Rows are ordered by the value Mistral Large 4 is now open weight published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.