Model profile
Llama 3.1 70B Instruct
Evidence summary
Safety. Llama 3.1 70B Instruct has an estimated Safety rank of #256; its 90% source-sensitivity interval is #156–#279. Its behavior-only rank is #239; company governance moves the combined estimate to #256. Published Safety evidence spans 23 eval lineages and 7 of 7 components. Its strongest relative result is FORTRESS (over_refusal_score, #3 of 59); its weakest is LLM Ethics Benchmark (score, #5 of 5).
Freedom. Llama 3.1 70B Instruct has an estimated Freedom rank of #150; its 90% source-sensitivity interval is #134–#293. Published Freedom evidence spans 16 eval lineages and 1 of 1 components. Its strongest relative result is FORTRESS (over_refusal_score, #3 of 59); its weakest is UGI Leaderboard — base-model willingness (willingness_adherence_score, #151 of 156).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗llama-3-1-instruct-70b
- Hugging Face ↗meta-llama/Meta-Llama-3.1-70B-Instruct
- NVIDIA ↗qc69jvmznzxy/llama-3_1-70b-instruct
- OpenRouter ↗meta-llama/llama-3.1-70b-instruct
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- 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 | #169 / 358 | ↓0.7821 | Source ↗official | |
| AbstentionBenchanswer_unknown_f1 | #13 / 20 | ↑0.8615 | Source ↗official | |
| AbstentionBenchfalse_premise_f1 | #10 / 20 | ↑0.676 | Source ↗official | |
| AbstentionBenchstale_f1 | #14 / 20 | ↑0.6032 | Source ↗official | |
| AbstentionBenchsubjective_f1 | #13 / 20 | ↑0.7183 | Source ↗official | |
| AbstentionBenchunderspecified_context_f1 | #5 / 20 | ↑0.6927 | Source ↗official | |
| AbstentionBenchunderspecified_intent_f1 | #15 / 20 | ↑0.702 | Source ↗official | |
| Agent-SafetyBenchcompromise_availability | #12 / 16 | ↑24 | Source ↗official | |
| Agent-SafetyBenchharmful_vulnerable_code | #10 / 16 | ↑29.6 | Source ↗official | |
| Agent-SafetyBenchleak_sensitive_information | #14 / 16 | ↑20 | Source ↗official | |
| Agent-SafetyBenchphysical_harm | #14 / 16 | ↑23.2 | Source ↗official | |
| Agent-SafetyBenchproduce_unsafe_information | #9 / 16 | ↑93.2 | Source ↗official | |
| Agent-SafetyBenchproperty_loss | #13 / 16 | ↑28.4 | Source ↗official | |
| Agent-SafetyBenchspread_unsafe_information | #11 / 16 | ↑10.8 | Source ↗official | |
| Agent-SafetyBenchviolate_law_ethics | #12 / 16 | ↑20.4 | Source ↗official | |
| AgentDrive Safety Compliancescr | #30 / 48 | ↑75 | Source ↗official | |
| AgentHarmharm_score | #5 / 12 | ↓14 | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #72 / 80 | ↑0.425 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #54 / 104 | ↓38.98 | Source ↗official | |
| Contextual MoralChoicehuman_agreement | #6 / 22 | ↑0.45 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #173 / 270 | ↑13.7 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #150 / 270 | ↑86.5 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #163 / 270 | ↑62.78 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #155 / 268 | ↑95.64 | Source ↗official | |
| FORTRESSaverage_risk_score | #42 / 60 | ↓44.18 | Source ↗official | |
| FORTRESSover_refusal_score | #3 / 59 | ↓1.09 | Source ↗official | |
| HELM Safetyanthropic_red_team | #74 / 80 | ↑0.932 | Source ↗official | |
| HELM Safetybbq | #23 / 80 | ↑0.954 | Source ↗official | |
| HELM Safetyharmbench | #74 / 80 | ↑0.469 | Source ↗official | |
| HELM Safetysimple_safety_tests | #72 / 80 | ↑0.925 | Source ↗official | |
| HELM Safetyxstest | #54 / 80 | ↑0.945 | Source ↗official | |
| Large-scale Moral Machine experiment on LLMshuman_choice_distance | #4 / 39 | ↓0.7398 | Source ↗official | |
| LLM Ethics Benchmarkscore | #5 / 5 | ↑75.8 | Source ↗official | |
| MedSafetyBenchmedical_safety_score | #7 / 30 | ↑96.39 | Source ↗official | |
| PandaBench JBB direct-request panelsafety_rate | #35 / 46 | ↑0.95 | Source ↗official | |
| PropensityBenchscore | #9 / 14 | ↓55.4 | Source ↗official | |
| SORRY-Benchavg | #38 / 51 | ↓0.39 | Source ↗official | |
| ThaiSafetyBenchsafety_score | #14 / 18 | ↑75.51 | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #14 / 25 | ↓8.6 | 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 |
|---|---|---|---|---|
| Agent-SafetyBenchproduce_unsafe_information | #8 / 16 | ↓93.2 | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #9 / 80 | ↓0.425 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #51 / 104 | ↑38.98 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #120 / 270 | ↓86.5 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #108 / 270 | ↓62.78 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #109 / 268 | ↓95.64 | Source ↗official | |
| FORTRESSaverage_risk_score | #19 / 60 | ↑44.18 | Source ↗official | |
| FORTRESSover_refusal_score | #3 / 59 | ↓1.09 | Source ↗official | |
| HELM Safetyanthropic_red_team | #7 / 80 | ↓0.932 | Source ↗official | |
| HELM Safetyharmbench | #7 / 80 | ↓0.469 | Source ↗official | |
| HELM Safetysimple_safety_tests | #9 / 80 | ↓0.925 | Source ↗official | |
| HELM Safetyxstest | #54 / 80 | ↑0.945 | Source ↗official | |
| MedSafetyBenchmedical_safety_score | #24 / 30 | ↓96.39 | Source ↗official | |
| PandaBench JBB direct-request panelsafety_rate | #8 / 46 | ↓0.95 | Source ↗official | |
| SORRY-Benchavg | #14 / 51 | ↑0.39 | Source ↗official | |
| SpeechMap model completioncomplete_pct | #81 / 181 | ↑61.7 | Source ↗official | |
| ThaiSafetyBenchsafety_score | #5 / 18 | ↓75.51 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_adherence_score | #151 / 156 | ↑0 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_direct_score | #79 / 156 | ↑3 | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #14 / 25 | ↓8.6 | 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.