Model profile
Llama 3 8B Instruct
Evidence summary
Safety. Llama 3 8B Instruct has an estimated Safety rank of #242; its 90% source-sensitivity interval is #156–#261. Its behavior-only rank is #225; company governance moves the combined estimate to #242. Published Safety evidence spans 16 eval lineages and 7 of 7 components. Its strongest relative result is Microsoft Phi Safety Panels (harmful_summarization, #1 of 10); its weakest is Contextual MoralChoice (human_agreement, #21 of 22).
Freedom. Llama 3 8B Instruct has an estimated Freedom rank of #447; its 90% source-sensitivity interval is #284–#519. Published Freedom evidence spans 13 eval lineages and 1 of 1 components. Its strongest relative result is Microsoft Phi Safety Panels (third_party_harm, #4 of 10); its weakest is Microsoft Phi Safety Panels (harmful_summarization, #10 of 10).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗llama-3-instruct-8b
- OpenRouter ↗meta-llama/llama-3-8b-instruct
- Official model page ↗Family-level model document · meta · first party
- Release source ↗direct research preview identity date
Safety evals
Rank and direction are specific to the Safety portfolio. Black marks the observed result; the grey dot marks the value implied by the Safety rank. Values stay on each source’s native scale.
| Eval / sub-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #282 / 358 | ↓0.8979 | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #35 / 80 | ↑0.709 | Source ↗official | |
| Contextual MoralChoicehuman_agreement | #21 / 22 | ↑0.3 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #122 / 270 | ↑17.19 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #159 / 270 | ↑85.83 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #77 / 270 | ↑89.16 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #130 / 268 | ↑96.45 | Source ↗official | |
| HELM Safetyanthropic_red_team | #41 / 80 | ↑0.988 | Source ↗official | |
| HELM Safetybbq | #70 / 80 | ↑0.765 | Source ↗official | |
| HELM Safetyharmbench | #43 / 80 | ↑0.727 | Source ↗official | |
| HELM Safetysimple_safety_tests | #29 / 80 | ↑0.993 | Source ↗official | |
| HELM Safetyxstest | #43 / 80 | ↑0.956 | Source ↗official | |
| JailBenchjailbreak_success_rate | #2 / 14 | ↓43.94 | Source ↗official | |
| Large-scale Moral Machine experiment on LLMshuman_choice_distance | #8 / 39 | ↓0.8324 | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_continuation | #6 / 10 | ↓0.013 | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_summarization | #1 / 10 | ↓0.082 | Source ↗official | |
| Microsoft Phi Safety Panelsjailbreak | #7 / 10 | ↓0.13 | Source ↗official | |
| Microsoft Phi Safety Panelsthird_party_harm | #7 / 10 | ↓0.373 | Source ↗official | |
| MT-JailBench CrescendoXsafety_score | #8 / 21 | ↑28.93 | Source ↗official | |
| MuPPET Contextual Privacymultiparty_contextual_privacy_score | #4 / 7 | ↑41.71 | Source ↗official | |
| OR-Benchover_refusal_rate | #15 / 25 | ↓69.4 | Source ↗official | |
| OR-Benchtoxic_acceptance_rate | #14 / 25 | ↓5 | Source ↗official | |
| S-Evalbase_en_overall | #8 / 22 | ↑69.1 | Source ↗official | |
| SORRY-Benchavg | #19 / 51 | ↓0.22 | Source ↗official |
Freedom evals
Rank and direction are specific to the Freedom portfolio. Black marks the observed result; the grey dot marks the value implied by the Freedom rank. Values stay on each source’s native scale.
| Eval / sub-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| AIRBench 2024 Safety Scenariossafety_scenarios | #46 / 80 | ↓0.709 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #111 / 270 | ↓85.83 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #194 / 270 | ↓89.16 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #139 / 268 | ↓96.45 | Source ↗official | |
| HELM Safetyanthropic_red_team | #38 / 80 | ↓0.988 | Source ↗official | |
| HELM Safetyharmbench | #38 / 80 | ↓0.727 | Source ↗official | |
| HELM Safetysimple_safety_tests | #50 / 80 | ↓0.993 | Source ↗official | |
| HELM Safetyxstest | #43 / 80 | ↑0.956 | Source ↗official | |
| JailBenchjailbreak_success_rate | #13 / 14 | ↑43.94 | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_continuation | #4 / 10 | ↑0.013 | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_summarization | #10 / 10 | ↑0.082 | Source ↗official | |
| Microsoft Phi Safety Panelsjailbreak | #4 / 10 | ↑0.13 | Source ↗official | |
| Microsoft Phi Safety Panelsthird_party_harm | #4 / 10 | ↑0.373 | Source ↗official | |
| MT-JailBench CrescendoXsafety_score | #14 / 21 | ↓28.93 | Source ↗official | |
| MuPPET Contextual Privacymultiparty_contextual_privacy_score | #4 / 7 | ↓41.71 | Source ↗official | |
| OR-Benchover_refusal_rate | #15 / 25 | ↓69.4 | Source ↗official | |
| OR-Benchtoxic_acceptance_rate | #12 / 25 | ↑5 | Source ↗official | |
| S-Evalbase_en_overall | #15 / 22 | ↓69.1 | Source ↗official | |
| SORRY-Benchavg | #33 / 51 | ↑0.22 | Source ↗official | |
| SpeechMap model completioncomplete_pct | #157 / 181 | ↑34.4 | Source ↗official |