Model profile
Deepseek V2.5
2024-09-05release date
1eval lineages
Evidence summary
Published evidence spans 1 evals and 2 of 7 behavior components. Its strongest relative result is Agent-SafetyBench (compromise_availability, #7 of 16); its weakest is Agent-SafetyBench (produce_unsafe_information, #13 of 16).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗deepseek-v2-5
- Hugging Face ↗deepseek-ai/DeepSeek-V2.5
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- Release source ↗direct research preview identity date
Published eval results
Rank is within that sub-eval. Black marks the observed result; the grey dot marks the value implied by the global rank. Values stay on each source’s native scale.
| Eval / sub-eval | Rank | Value | Distribution | Better | Source |
|---|---|---|---|---|---|
| Agent-SafetyBenchcompromise_availability | #7 / 16 | 33.2 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchharmful_vulnerable_code | #9 / 16 | 30.4 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchleak_sensitive_information | #9 / 16 | 31.2 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchphysical_harm | #8 / 16 | 34.4 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchproduce_unsafe_information | #13 / 16 | 76.8 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchproperty_loss | #10 / 16 | 36.8 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchspread_unsafe_information | #12 / 16 | 8.8 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchviolate_law_ethics | #11 / 16 | 22 | ↑ higher | Source ↗official |