Model profile
Grok 3
Evidence summary
Grok 3 has an estimated overall rank of #163; its 90% source-sensitivity interval is #52–#228. Its behavior-only rank is #167; company governance moves the combined estimate to #163. Published evidence spans 17 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 | Better | Source |
|---|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #194 / 311 | 0.8461 | ↓ lower | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #64 / 80 | 0.513 | ↑ higher | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #104 / 105 | 84.15 | ↓ lower | Source ↗official | |
| Confabulationsconfabulation_rate | #19 / 52 | 17.82 | ↓ lower | Source ↗official | |
| Emergent Collusionhigh_illegality_game_rate | #1 / 13 | 0.19 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #95 / 260 | 19.38 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #139 / 260 | 88 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #221 / 260 | 40.56 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #253 / 258 | 61.75 | ↑ higher | Source ↗official | |
| HELM Safetyanthropic_red_team | #72 / 80 | 0.955 | ↑ higher | Source ↗official | |
| HELM Safetybbq | #41 / 80 | 0.936 | ↑ higher | Source ↗official | |
| HELM Safetyharmbench | #75 / 80 | 0.453 | ↑ higher | Source ↗official | |
| HELM Safetysimple_safety_tests | #62 / 80 | 0.968 | ↑ higher | Source ↗official | |
| HELM Safetyxstest | #33 / 80 | 0.964 | ↑ higher | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #5 / 54 | 29.14 | ↑ higher | Source ↗official | |
| PHAREbias_resistance_diagnostic | #65 / 66 | 0.2324 | ↑ higher | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #36 / 70 | 0.7512 | ↑ higher | Source ↗official | |
| PHAREharm_resistance_diagnostic | #51 / 70 | 0.8968 | ↑ higher | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #66 / 67 | 0.2577 | ↑ higher | Source ↗official | |
| SOSBenchbiology_pvr | #23 / 23 | 0.894 | ↓ lower | Source ↗official | |
| SOSBenchchemistry_pvr | #18 / 23 | 0.638 | ↓ lower | Source ↗official | |
| SOSBenchmedicine_pvr | #21 / 23 | 0.86 | ↓ lower | Source ↗official | |
| SOSBenchpharmacology_pvr | #21 / 23 | 0.954 | ↓ lower | Source ↗official | |
| SOSBenchphysics_pvr | #20 / 23 | 0.804 | ↓ lower | Source ↗official | |
| SOSBenchpsychology_pvr | #23 / 23 | 0.89 | ↓ lower | Source ↗official | |
| SpeciesismBenchexplicit_speciesism_scale | #6 / 7 | 2.493 | ↓ lower | Source ↗official | |
| SpeciesismBenchmorally_wrong_rate | #4 / 8 | 38.78 | ↑ higher | Source ↗official | |
| SpeciesismBenchspeciesism_recognition_rate | #3 / 8 | 89.13 | ↑ higher | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #98 / 102 | 6.255 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #93 / 102 | 5.025 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #84 / 102 | 5.05 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #93 / 102 | 2.865 | ↓ lower | Source ↗official | |
| TukaBenchafri_jbb_cultural_asr | #6 / 6 | 35.6 | ↓ lower | Source ↗official | |
| TukaBenchafri_jbb_harm_asr | #6 / 6 | 29.3 | ↓ lower | Source ↗official | |
| TukaBenchafrijail_mono_asr | #6 / 6 | 30.2 | ↓ lower | 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.