# Safety Evaluation on HarmBench and SafeChain (Safety-Tuned) Source: [Safety Evaluation on HarmBench and SafeChain (Safety-Tuned)]() Collection index: [llms.txt]() Data updated: 2026-02-27T10:52:02.905Z Evaluation of model safety performance combining results from the HarmBench and SafeChain datasets under a specific Safety-Tuned protocol. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Safety | | Dataset (including any stated split) | HarmBench, SafeChain Safety-Tuned | | Metric | Safety-Tuned Score | | Unit | SCALAR | | Direction | Higher is better | | Results loaded | 18 of 18 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Safety-Tuned Score](): Higher is better; unit: SCALAR. - [HarmBench Score](): Higher is better; unit: SCALAR. - [SafeChain Score](): Higher is better; unit: SCALAR. ## Recorded results | Method | Safety-Tuned Score | Result date | Report paper | | --- | --- | --- | --- | | LED | 2.41 | Not recorded | 69647fe953fdd5c36174478b | | Task Arithmetic | 2.31 | Not recorded | 69647fe953fdd5c36174478b | | Linear | 2.08 | Not recorded | 69647fe953fdd5c36174478b | | Ties | 1.95 | Not recorded | 69647fe953fdd5c36174478b | | Reasoning | 1.91 | Not recorded | 69647fe953fdd5c36174478b | | FuseLLM | 1.73 | Not recorded | 69647fe953fdd5c36174478b | | Reasoning | 1.72 | Not recorded | 69647fe953fdd5c36174478b | | LED | 1.57 | Not recorded | 69647fe953fdd5c36174478b | | DARE | 1.54 | Not recorded | 69647fe953fdd5c36174478b | | Task Arithmetic | 1.4 | Not recorded | 69647fe953fdd5c36174478b | | TIES | 1.32 | Not recorded | 69647fe953fdd5c36174478b | | Linear | 1.28 | Not recorded | 69647fe953fdd5c36174478b | | DARE | 1.28 | Not recorded | 69647fe953fdd5c36174478b | | Safety | 1.2 | Not recorded | 69647fe953fdd5c36174478b | | ReasonAny | 1.2 | Not recorded | 69647fe953fdd5c36174478b | | ReasonAny | 1.19 | Not recorded | 69647fe953fdd5c36174478b | | Safety | 1.18 | Not recorded | 69647fe953fdd5c36174478b | | FuseLLM | 0.87 | Not recorded | 69647fe953fdd5c36174478b | ### LED Safety-Tuned Score: 2.41. Result ID: 69a1774c35dfcabb00e7374a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B"}. This record does not assert independent reproduction. ### Task Arithmetic Safety-Tuned Score: 2.31. Result ID: 69a1774c35dfcabb00e73746. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B"}. This record does not assert independent reproduction. ### Linear Safety-Tuned Score: 2.08. Result ID: 69a1774c35dfcabb00e73745. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B"}. This record does not assert independent reproduction. ### Ties Safety-Tuned Score: 1.95. Result ID: 69a1774c35dfcabb00e73747. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B"}. This record does not assert independent reproduction. ### Reasoning Safety-Tuned Score: 1.91. Result ID: 69a1774c35dfcabb00e73744. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B"}. This record does not assert independent reproduction. ### FuseLLM Safety-Tuned Score: 1.73. Result ID: 69a1774c35dfcabb00e7388d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merging Method":"FuseLLM"}. This record does not assert independent reproduction. ### Reasoning Safety-Tuned Score: 1.72. Result ID: 69a1774c35dfcabb00e73888. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Family":"Qwen2.5-32B","Role":"Source Model (QwQ-32B-Preview)"}. This record does not assert independent reproduction. ### LED Safety-Tuned Score: 1.57. Result ID: 69a1774c35dfcabb00e7388e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merging Method":"LED"}. This record does not assert independent reproduction. ### DARE Safety-Tuned Score: 1.54. Result ID: 69a1774c35dfcabb00e73748. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B"}. This record does not assert independent reproduction. ### Task Arithmetic Safety-Tuned Score: 1.4. Result ID: 69a1774c35dfcabb00e7388a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merging Method":"Task Arithmetic"}. This record does not assert independent reproduction. ### TIES Safety-Tuned Score: 1.32. Result ID: 69a1774c35dfcabb00e7388b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merging Method":"TIES"}. This record does not assert independent reproduction. ### Linear Safety-Tuned Score: 1.28. Result ID: 69a1774c35dfcabb00e73889. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Merging Method":"Linear"}. This record does not assert independent reproduction. ### DARE Safety-Tuned Score: 1.28. Result ID: 69a1774c35dfcabb00e7388c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merging Method":"DARE"}. This record does not assert independent reproduction. ### Safety Safety-Tuned Score: 1.2. Result ID: 69a1774c35dfcabb00e73887. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Family":"Qwen2.5-32B","Role":"Source Model (Safety-Instruct)"}. This record does not assert independent reproduction. ### ReasonAny Safety-Tuned Score: 1.2. Result ID: 69a1774c35dfcabb00e7388f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merging Method":"ReasonAny"}. This record does not assert independent reproduction. ### ReasonAny Safety-Tuned Score: 1.19. Result ID: 69a1774c35dfcabb00e7374b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B"}. This record does not assert independent reproduction. ### Safety Safety-Tuned Score: 1.18. Result ID: 69a1774c35dfcabb00e73743. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B"}. This record does not assert independent reproduction. ### FuseLLM Safety-Tuned Score: 0.87. Result ID: 69a1774c35dfcabb00e73749. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.