Model profile
Grok 3
Evidence summary
Grok 3 has an estimated overall rank of #181; its 90% source-sensitivity interval is #76–#251. Its behavior-only rank is #185; company governance moves the combined estimate to #181. Published evidence spans 18 evals and 7 of 7 behavior components. Its strongest relative result is Emergent Collusion (high_illegality_game_rate, #1 of 13); its weakest is SOSBench (biology_pvr, #23 of 23).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗grok-3
- OpenRouter ↗x-ai/grok-3
- Official model documentation ↗Family-level model document · xai · first party
- 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 | Source |
|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #222 / 328 | ↓0.8732 | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #64 / 80 | ↑0.513 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #103 / 104 | ↓84.15 | Source ↗official | |
| Confabulationsconfabulation_rate | #19 / 52 | ↓17.82 | Source ↗official | |
| Emergent Collusionhigh_illegality_game_rate | #1 / 13 | ↓0.19 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #86 / 241 | ↑19.38 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #126 / 241 | ↑88 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #202 / 241 | ↑40.56 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #234 / 239 | ↑61.75 | Source ↗official | |
| HELM Safetyanthropic_red_team | #72 / 80 | ↑0.955 | Source ↗official | |
| HELM Safetybbq | #41 / 80 | ↑0.936 | Source ↗official | |
| HELM Safetyharmbench | #75 / 80 | ↑0.453 | Source ↗official | |
| HELM Safetysimple_safety_tests | #62 / 80 | ↑0.968 | Source ↗official | |
| HELM Safetyxstest | #33 / 80 | ↑0.964 | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #5 / 54 | ↑29.14 | Source ↗official | |
| PHAREbias_resistance_diagnostic | #65 / 66 | ↑0.2324 | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #36 / 70 | ↑0.7512 | Source ↗official | |
| PHAREharm_resistance_diagnostic | #51 / 70 | ↑0.8968 | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #66 / 67 | ↑0.2577 | Source ↗official | |
| SOSBenchbiology_pvr | #23 / 23 | ↓0.894 | Source ↗official | |
| SOSBenchchemistry_pvr | #18 / 23 | ↓0.638 | Source ↗official | |
| SOSBenchmedicine_pvr | #21 / 23 | ↓0.86 | Source ↗official | |
| SOSBenchpharmacology_pvr | #21 / 23 | ↓0.954 | Source ↗official | |
| SOSBenchphysics_pvr | #20 / 23 | ↓0.804 | Source ↗official | |
| SOSBenchpsychology_pvr | #23 / 23 | ↓0.89 | Source ↗official | |
| SpeciesismBenchexplicit_speciesism_scale | #6 / 7 | ↓2.493 | Source ↗official | |
| SpeciesismBenchmorally_wrong_rate | #4 / 8 | ↑38.78 | Source ↗official | |
| SpeciesismBenchspeciesism_recognition_rate | #3 / 8 | ↑89.13 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #109 / 113 | ↑6.255 | Source ↗official | |
| SpeciEvalland_animal_4ns | #104 / 113 | ↓5.025 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #94 / 113 | ↓5.05 | Source ↗official | |
| SpeciEvalspeciesism | #104 / 113 | ↓2.865 | Source ↗official | |
| TukaBenchafri_jbb_cultural_asr | #6 / 6 | ↓35.6 | Source ↗official | |
| TukaBenchafri_jbb_harm_asr | #6 / 6 | ↓29.3 | Source ↗official | |
| TukaBenchafrijail_mono_asr | #6 / 6 | ↓30.2 | Source ↗official | |
| Vectara HHEM Factual Consistencyfactual_consistency_rate | #21 / 94 | ↑94.2 | 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.