← Models

Model profile

Llama 3.1 70B Instruct

Metadeveloper
2024-07-23release date
#187 / 267overall rank
19eval lineages

Evidence summary

Llama 3.1 70B Instruct has an estimated overall rank of #187; its 90% source-sensitivity interval is #53–#213. Its behavior-only rank is #177; company governance moves the combined estimate to #187. Published evidence spans 19 evals and 7 of 7 behavior components. Its strongest relative result is FORTRESS (over_refusal_score, #3 of 46); its weakest is LLM Ethics Benchmark (score, #5 of 5).

Compare this model

Only models sharing at least one published sub-eval are listed.

Official and reference links

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-evalRankValueDistributionBetterSource
AA-Omnisciencehallucination_rate#117 / 3110.7472↓ lowerSource ↗official
AbstentionBenchanswer_unknown_f1#13 / 200.8615↑ higherSource ↗official
AbstentionBenchfalse_premise_f1#10 / 200.676↑ higherSource ↗official
AbstentionBenchstale_f1#14 / 200.6032↑ higherSource ↗official
AbstentionBenchsubjective_f1#13 / 200.7183↑ higherSource ↗official
AbstentionBenchunderspecified_context_f1#5 / 200.6927↑ higherSource ↗official
AbstentionBenchunderspecified_intent_f1#15 / 200.702↑ higherSource ↗official
Agent-SafetyBenchcompromise_availability#12 / 1624↑ higherSource ↗official
Agent-SafetyBenchharmful_vulnerable_code#10 / 1629.6↑ higherSource ↗official
Agent-SafetyBenchleak_sensitive_information#14 / 1620↑ higherSource ↗official
Agent-SafetyBenchphysical_harm#14 / 1623.2↑ higherSource ↗official
Agent-SafetyBenchproduce_unsafe_information#9 / 1693.2↑ higherSource ↗official
Agent-SafetyBenchproperty_loss#13 / 1628.4↑ higherSource ↗official
Agent-SafetyBenchspread_unsafe_information#11 / 1610.8↑ higherSource ↗official
Agent-SafetyBenchviolate_law_ethics#12 / 1620.4↑ higherSource ↗official
AgentHarmharm_score#5 / 1214↓ lowerSource ↗official
AIRBench 2024 Safety Scenariossafety_scenarios#72 / 800.425↑ higherSource ↗official
Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate#55 / 10538.98↓ lowerSource ↗official
Contextual MoralChoicehuman_agreement#6 / 220.45↑ higherSource ↗official
FORTRESSaverage_risk_score#31 / 4944.18↓ lowerSource ↗official
FORTRESSover_refusal_score#3 / 461.09↓ lowerSource ↗official
HELM Safetyanthropic_red_team#74 / 800.932↑ higherSource ↗official
HELM Safetybbq#23 / 800.954↑ higherSource ↗official
HELM Safetyharmbench#74 / 800.469↑ higherSource ↗official
HELM Safetysimple_safety_tests#72 / 800.925↑ higherSource ↗official
HELM Safetyxstest#54 / 800.945↑ higherSource ↗official
Large-scale Moral Machine experiment on LLMshuman_choice_distance#4 / 390.7398↓ lowerSource ↗official
LLM Ethics Benchmarkscore#5 / 575.8↑ higherSource ↗official
PandaBench JBB direct-request panelsafety_rate#35 / 460.95↑ higherSource ↗official
PropensityBenchscore#9 / 1455.4↓ lowerSource ↗official
SORRY-Benchavg#38 / 510.39↓ lowerSource ↗official
VETO Misfired Alignmentmisfired_alignment_rate_pct#14 / 258.6↓ lowerSource ↗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.

UGI Political Values

DimensionValueDistribution
Political Lean-17.6
Government47.5
Diplomacy66.1
Economy42.5
Society62

ValueCompass

DimensionValueDistribution
Universalism60.2
Self-direction41.6
Care / Harm17.3
Fairness / Cheating17.5
Ethical88.1