Model profile
Llama 3 70B Instruct
Evidence summary
Llama 3 70B Instruct has an estimated overall rank of #185; its 90% source-sensitivity interval is #93–#220. Its behavior-only rank is #171; company governance moves the combined estimate to #185. Published evidence spans 13 evals and 7 of 7 behavior components. Its strongest relative result is Enkrypt AI Safety Leaderboard (toxicity_attack_non_success_rate, #1 of 258); its weakest is AgentDojo (utility_under_attack, #15 of 15).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗llama-3-instruct-70b
- OpenRouter ↗meta-llama/llama-3-70b-instruct
- 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 | Better | Source |
|---|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #209 / 311 | 0.865 | ↓ lower | Source ↗official | |
| AgentDojotargeted_attack_success_rate | #11 / 15 | 0.256 | ↓ lower | Source ↗official | |
| AgentDojoutility_under_attack | #15 / 15 | 0.1828 | ↑ higher | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #45 / 80 | 0.646 | ↑ higher | Source ↗official | |
| CASE-Benchagreement_accuracy | #3 / 7 | 84.44 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #92 / 260 | 19.64 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #168 / 260 | 86.33 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #63 / 260 | 87.78 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #1 / 258 | 100 | ↑ higher | Source ↗official | |
| HELM Safetyanthropic_red_team | #66 / 80 | 0.967 | ↑ higher | Source ↗official | |
| HELM Safetybbq | #53 / 80 | 0.91 | ↑ higher | Source ↗official | |
| HELM Safetyharmbench | #57 / 80 | 0.64 | ↑ higher | Source ↗official | |
| HELM Safetysimple_safety_tests | #32 / 80 | 0.99 | ↑ higher | Source ↗official | |
| HELM Safetyxstest | #29 / 80 | 0.968 | ↑ higher | Source ↗official | |
| Large-scale Moral Machine experiment on LLMshuman_choice_distance | #5 / 39 | 0.7475 | ↓ lower | Source ↗official | |
| OR-Benchover_refusal_rate | #9 / 25 | 37.7 | ↓ lower | Source ↗official | |
| OR-Benchtoxic_acceptance_rate | #22 / 25 | 21.3 | ↓ lower | Source ↗official | |
| S-Evalbase_en_overall | #15 / 22 | 54.7 | ↑ higher | Source ↗official | |
| SORRY-Benchavg | #31 / 51 | 0.35 | ↓ lower | Source ↗official |
Values evaluations
Descriptive values and political-framing results are separate from safety/ethics ranks. Each strip shows the evaluation’s observed model range; its endpoint labels state what lower and higher values mean.