Model profile
Llama 2 13B Chat
Metadeveloper
2023-07-18release date
#197 / 309overall rank
9eval lineages
Evidence summary
Llama 2 13B Chat has an estimated overall rank of #197; its 90% source-sensitivity interval is #87–#247. Its behavior-only rank is #186; company governance moves the combined estimate to #197. Published evidence spans 9 evals and 6 of 7 behavior components. Its strongest relative result is COMPL-AI AI-Identity Disclosure (score, #1 of 14); its weakest is SafetyBench (OFF, #20 of 21).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- OpenRouter ↗meta-llama/llama-2-13b-chat
- Official model page ↗Family-level model document · meta · first party
- 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 | Source |
|---|---|---|---|---|
| COMPL-AI AI-Identity Disclosurescore | #1 / 14 | ↑1 | Source ↗official | |
| COMPL-AI LLM RuLES Multi-Turn Rule Followingscore | #9 / 14 | ↑0.3652 | Source ↗official | |
| COMPL-AI TensorTrust Goal-Hijacking Resistancescore | #11 / 13 | ↑0.4175 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #195 / 241 | ↑11.11 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #51 / 241 | ↑91.83 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #53 / 241 | ↑88.33 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #20 / 239 | ↑99.55 | Source ↗official | |
| HarmBenchdr | #4 / 28 | ↓2.8 | Source ↗official | |
| JailBenchjailbreak_success_rate | #6 / 14 | ↓55.39 | Source ↗official | |
| OR-Benchover_refusal_rate | #20 / 25 | ↓91 | Source ↗official | |
| OR-Benchtoxic_acceptance_rate | #2 / 25 | ↓0.3 | Source ↗official | |
| S-Evalbase_en_overall | #2 / 22 | ↑85.1 | Source ↗official | |
| SafetyBenchEM | #17 / 21 | ↑54.6 | Source ↗official | |
| SafetyBenchIA | #15 / 21 | ↑68.5 | Source ↗official | |
| SafetyBenchMH | #15 / 21 | ↑73.6 | Source ↗official | |
| SafetyBenchOFF | #20 / 21 | ↑48.4 | Source ↗official | |
| SafetyBenchPH | #16 / 21 | ↑60.7 | Source ↗official | |
| SafetyBenchPP | #16 / 21 | ↑70.1 | Source ↗official | |
| SafetyBenchUB | #5 / 21 | ↑66.3 | Source ↗official | |
| SALAD-Benchattack_enhanced_human_autonomy_integrity | #4 / 33 | ↑62.72 | Source ↗official | |
| SALAD-Benchattack_enhanced_information_safety_harms | #4 / 33 | ↑71.01 | Source ↗official | |
| SALAD-Benchattack_enhanced_malicious_use | #5 / 33 | ↑62.56 | Source ↗official | |
| SALAD-Benchattack_enhanced_misinformation_harms | #4 / 33 | ↑70.23 | Source ↗official | |
| SALAD-Benchattack_enhanced_representation_toxicity | #5 / 33 | ↑65.85 | Source ↗official | |
| SALAD-Benchattack_enhanced_socioeconomic_harms | #4 / 33 | ↑68.4 | Source ↗official | |
| SALAD-Benchbase_human_autonomy_integrity | #4 / 33 | ↑98.43 | Source ↗official | |
| SALAD-Benchbase_information_safety_harms | #16 / 33 | ↑94.92 | Source ↗official | |
| SALAD-Benchbase_malicious_use | #7 / 33 | ↑97.98 | Source ↗official | |
| SALAD-Benchbase_misinformation_harms | #4 / 33 | ↑97.64 | Source ↗official | |
| SALAD-Benchbase_representation_toxicity | #4 / 33 | ↑95.71 | Source ↗official | |
| SALAD-Benchbase_socioeconomic_harms | #13 / 33 | ↑91.19 | Source ↗official | |
| SALAD-Benchmcq_human_autonomy_integrity | #26 / 33 | ↑10.28 | Source ↗official | |
| SALAD-Benchmcq_information_safety_harms | #23 / 33 | ↑17.5 | Source ↗official | |
| SALAD-Benchmcq_malicious_use | #26 / 33 | ↑7.885 | Source ↗official | |
| SALAD-Benchmcq_misinformation_harms | #26 / 33 | ↑10.95 | Source ↗official | |
| SALAD-Benchmcq_representation_toxicity | #26 / 33 | ↑8.229 | Source ↗official | |
| SALAD-Benchmcq_socioeconomic_harms | #26 / 33 | ↑12.78 | Source ↗official | |
| SORRY-Benchavg | #15 / 51 | ↓0.15 | Source ↗official |