Model profile
Llama 4 Maverick
Metadeveloper
2025-04-05release date
#165 / 267overall rank
28eval lineages
Evidence summary
Llama 4 Maverick has an estimated overall rank of #165; its 90% source-sensitivity interval is #104–#201. Its behavior-only rank is #160; company governance moves the combined estimate to #165. Published evidence spans 28 evals and 7 of 7 behavior components. Its strongest relative result is Anthropic Agentic Misalignment — blackmail (misaligned_action_rate_pct, #1 of 16); its weakest is SpeciesismBench (explicit_speciesism_scale, #7 of 7).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗llama-4-maverick
- Hugging Face ↗meta-llama/Llama-4-Maverick-17B-128E-Instruct
- OpenRouter ↗meta-llama/llama-4-maverick
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- 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 | #224 / 311 | 0.8732 | ↓ lower | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #37 / 80 | 0.686 | ↑ higher | Source ↗official | |
| Alignment Leaderboardcorrigibility | #15 / 24 | 4.175 | ↑ higher | Source ↗official | |
| Alignment Leaderboardhonesty | #14 / 24 | 3.581 | ↑ higher | Source ↗official | |
| Alignment Leaderboardnon_manipulation | #17 / 24 | 3.084 | ↑ higher | Source ↗official | |
| Alignment Leaderboardrobustness | #9 / 24 | 3.867 | ↑ higher | Source ↗official | |
| Alignment Leaderboardsafety | #18 / 24 | 3.567 | ↑ higher | Source ↗official | |
| Alignment Leaderboardscheming | #19 / 24 | 3.422 | ↑ higher | Source ↗official | |
| Anthropic Agentic Misalignment — blackmailmisaligned_action_rate_pct | #1 / 16 | 0 | ↓ lower | Source ↗official | |
| Anthropic Agentic Misalignment — corporate espionagemisaligned_action_rate_pct | #7 / 16 | 44 | ↓ lower | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #53 / 105 | 0.28 | ↑ higher | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #73 / 105 | 56.23 | ↓ lower | Source ↗official | |
| Confabulationsconfabulation_rate | #34 / 52 | 28.22 | ↓ lower | Source ↗official | |
| DystopiaBenchbasaglia_score | #27 / 50 | 64.93 | ↓ lower | Source ↗official | |
| DystopiaBenchbaudrillard_score | #36 / 50 | 65.83 | ↓ lower | Source ↗official | |
| DystopiaBenchhuxley_score | #27 / 50 | 72.8 | ↓ lower | Source ↗official | |
| DystopiaBenchlaguardia_score | #25 / 50 | 66.73 | ↓ lower | Source ↗official | |
| DystopiaBenchorwell_score | #23 / 50 | 68.6 | ↓ lower | Source ↗official | |
| DystopiaBenchpetrov_score | #18 / 50 | 68.63 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #222 / 260 | 10.34 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #71 / 260 | 91.17 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #88 / 260 | 82.22 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #92 / 258 | 97.82 | ↑ higher | Source ↗official | |
| FORTRESSaverage_risk_score | #28 / 49 | 40.09 | ↓ lower | Source ↗official | |
| FORTRESSover_refusal_score | #24 / 46 | 4.95 | ↓ lower | Source ↗official | |
| HELM Safetyanthropic_red_team | #54 / 80 | 0.982 | ↑ higher | Source ↗official | |
| HELM Safetybbq | #43 / 80 | 0.93 | ↑ higher | Source ↗official | |
| HELM Safetyharmbench | #52 / 80 | 0.661 | ↑ higher | Source ↗official | |
| HELM Safetysimple_safety_tests | #29 / 80 | 0.993 | ↑ higher | Source ↗official | |
| HELM Safetyxstest | #32 / 80 | 0.965 | ↑ higher | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #50 / 54 | 23.02 | ↑ higher | Source ↗official | |
| LiveSecBenchethics | #37 / 43 | 22.55 | ↑ higher | Source ↗official | |
| LiveSecBenchfactuality | #35 / 43 | 29.74 | ↑ higher | Source ↗official | |
| LiveSecBenchlegality | #30 / 43 | 27.05 | ↑ higher | Source ↗official | |
| LiveSecBenchprivacy | #35 / 43 | 26.17 | ↑ higher | Source ↗official | |
| LiveSecBenchpsychological_health | #29 / 43 | 35.37 | ↑ higher | Source ↗official | |
| Manager Coercion Benchcoercion_ladder_depth | #15 / 31 | 8.333 | ↓ lower | Source ↗self run | |
| MASKlying_probability_pct | #41 / 53 | 50.27 | ↓ lower | Source ↗official | |
| ODCV-Benchaverage_severity | #10 / 12 | 1.682 | ↓ lower | Source ↗official | |
| ODCV-Benchmisalignment_rate | #2 / 12 | 16.22 | ↓ lower | Source ↗official | |
| PHAREbias_resistance_diagnostic | #5 / 66 | 0.7365 | ↑ higher | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #43 / 70 | 0.7147 | ↑ higher | Source ↗official | |
| PHAREharm_resistance_diagnostic | #53 / 70 | 0.8925 | ↑ higher | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #34 / 67 | 0.4899 | ↑ higher | Source ↗official | |
| Shelleducation_jsr | #10 / 14 | 0.696 | ↓ lower | Source ↗official | |
| Shellfinance_jsr | #11 / 14 | 0.716 | ↓ lower | Source ↗official | |
| Shellmanagement_jsr | #11 / 14 | 0.844 | ↓ lower | Source ↗official | |
| SOSBenchbiology_pvr | #7 / 23 | 0.288 | ↓ lower | Source ↗official | |
| SOSBenchchemistry_pvr | #7 / 23 | 0.238 | ↓ lower | Source ↗official | |
| SOSBenchmedicine_pvr | #6 / 23 | 0.426 | ↓ lower | Source ↗official | |
| SOSBenchpharmacology_pvr | #8 / 23 | 0.652 | ↓ lower | Source ↗official | |
| SOSBenchphysics_pvr | #8 / 23 | 0.24 | ↓ lower | Source ↗official | |
| SOSBenchpsychology_pvr | #6 / 23 | 0.242 | ↓ lower | Source ↗official | |
| SpeciesismBenchexplicit_speciesism_scale | #7 / 7 | 3.3 | ↓ lower | Source ↗official | |
| SpeciesismBenchmorally_wrong_rate | #2 / 8 | 42.67 | ↑ higher | Source ↗official | |
| SpeciesismBenchspeciesism_recognition_rate | #4 / 8 | 88.97 | ↑ higher | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #19 / 102 | 6.97 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #69 / 102 | 4.72 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #70 / 102 | 4.9 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #87 / 102 | 2.65 | ↓ lower | Source ↗official | |
| TrustLLM contemporary collapsed applicationtrustllm | #7 / 8 | 0.6 | ↑ higher | Source ↗official | |
| TukaBenchafri_jbb_cultural_asr | #4 / 6 | 22.4 | ↓ lower | Source ↗official | |
| TukaBenchafri_jbb_harm_asr | #4 / 6 | 16.9 | ↓ lower | Source ↗official | |
| TukaBenchafrijail_mono_asr | #4 / 6 | 22 | ↓ lower | Source ↗official | |
| Vigil Mental Health Safetyoverall_score | #19 / 23 | 28 | ↑ higher | 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.