Model profile
Llama 3.2 3B Instruct
Evidence summary
Safety. Llama 3.2 3B Instruct has an estimated Safety rank of #329; its 90% source-sensitivity interval is #269–#331. Its behavior-only rank is #319; company governance moves the combined estimate to #329. Published Safety evidence spans 8 eval lineages and 5 of 7 components. Its strongest relative result is Open LLM Safety Index (strongreject_safety_rate, #5 of 21); its weakest is UAVBench safety-critical decision recognition (ethical_safety_critical_accuracy, #27 of 27).
Freedom. Llama 3.2 3B Instruct has an estimated Freedom rank of #260; its 90% source-sensitivity interval is #141–#480. Published Freedom evidence spans 6 eval lineages and 1 of 1 components. Its strongest relative result is BlueBench AttaQ-100 (attaq_harmlessness_reward_pct, #6 of 18); its weakest is UGI Leaderboard — base-model willingness (willingness_adherence_score, #144 of 156).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗llama-3-2-instruct-3b
- Hugging Face ↗meta-llama/Llama-3.2-3B-Instruct
- NVIDIA ↗qc69jvmznzxy/llama-3.2-3b-instruct
- OpenRouter ↗meta-llama/llama-3.2-3b-instruct
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- Release source ↗direct research preview identity date
Finetunes
Direct finetunes linked to this canonical base model. Quantized and repackaged derivatives are excluded.
huihui-ai/Llama-3.2-3B-Instruct-abliterated ↗Direct non-quantized finetune documented by Hugging Face metadata
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 |
|---|---|---|---|---|
| AgentDrive Safety Compliancescr | #44 / 48 | ↑43.75 | Source ↗official | |
| BlueBench AttaQ-100attaq_harmlessness_reward_pct | #13 / 18 | ↑83.37 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #57 / 104 | ↓40.58 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #203 / 248 | ↑11.11 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #160 / 248 | ↑85.67 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #77 / 248 | ↑85 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #146 / 246 | ↑95.5 | Source ↗official | |
| Large-scale Moral Machine experiment on LLMshuman_choice_distance | #34 / 39 | ↓1.519 | Source ↗official | |
| Open LLM Safety Indexjailbreakbench_safety_rate | #14 / 21 | ↑0.2667 | Source ↗official | |
| Open LLM Safety Indexstrongreject_safety_rate | #5 / 21 | ↑0.6667 | Source ↗official | |
| PandaBench JBB direct-request panelsafety_rate | #14 / 46 | ↑0.99 | Source ↗official | |
| UAVBench safety-critical decision recognitionethical_safety_critical_accuracy | #27 / 27 | ↑0.475 | 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 |
|---|---|---|---|---|
| BlueBench AttaQ-100attaq_harmlessness_reward_pct | #6 / 18 | ↓83.37 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #48 / 104 | ↑40.58 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #89 / 248 | ↓85.67 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #170 / 248 | ↓85 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #101 / 246 | ↓95.5 | Source ↗official | |
| Open LLM Safety Indexjailbreakbench_safety_rate | #7 / 21 | ↓0.2667 | Source ↗official | |
| Open LLM Safety Indexstrongreject_safety_rate | #17 / 21 | ↓0.6667 | Source ↗official | |
| PandaBench JBB direct-request panelsafety_rate | #24 / 46 | ↓0.99 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_adherence_score | #144 / 156 | ↑0.5 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_direct_score | #113 / 156 | ↑2 | 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.