Model profile
Grok 4
xAIdeveloper
2025-07-09release date
#105 / 267overall rank
27eval lineages
Evidence summary
Grok 4 has an estimated overall rank of #105; its 90% source-sensitivity interval is #47–#164. Its behavior-only rank is #109; company governance moves the combined estimate to #105. Published evidence spans 27 evals and 7 of 7 behavior components. Its strongest relative result is Confabulations (confabulation_rate, #3 of 52); its weakest is Emergent Collusion (high_illegality_game_rate, #13 of 13).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗grok-4
- OpenRouter ↗x-ai/grok-4
- 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 | #79 / 311 | 0.642 | ↓ lower | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #69 / 80 | 0.444 | ↑ higher | Source ↗official | |
| BrokenMathsycophancy | #5 / 9 | 43.4 | ↓ lower | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #29 / 43 | 85.9 | ↓ lower | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #20 / 48 | 64.2 | ↓ lower | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #25 / 49 | 56.4 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmachiavelli | #35 / 45 | 93.4 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmask | #35 / 51 | 41.7 | ↓ lower | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #18 / 32 | 49.5 | ↓ lower | Source ↗official | |
| CAIS Risk Indextextquests_harm | #46 / 48 | 24.3 | ↓ lower | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #39 / 105 | 21.78 | ↓ lower | Source ↗official | |
| Confabulationsconfabulation_rate | #3 / 52 | 3.96 | ↓ lower | Source ↗official | |
| DystopiaBenchbasaglia_score | #28 / 50 | 65.3 | ↓ lower | Source ↗official | |
| DystopiaBenchbaudrillard_score | #25 / 50 | 57.7 | ↓ lower | Source ↗official | |
| DystopiaBenchhuxley_score | #30 / 50 | 73.17 | ↓ lower | Source ↗official | |
| DystopiaBenchlaguardia_score | #15 / 50 | 59.07 | ↓ lower | Source ↗official | |
| DystopiaBenchorwell_score | #25 / 50 | 68.63 | ↓ lower | Source ↗official | |
| DystopiaBenchpetrov_score | #31 / 50 | 74.37 | ↓ lower | Source ↗official | |
| Emergent Collusionhigh_illegality_game_rate | #13 / 13 | 0.75 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #195 / 260 | 12.4 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #230 / 260 | 76.83 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #239 / 260 | 33.89 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #155 / 258 | 95.45 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa1_qualified_rate | #6 / 18 | 81.99 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa2_qualified_rate | #7 / 18 | 80.3 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa3_qualified_rate | #12 / 18 | 86.36 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa4_qualified_rate | #10 / 18 | 87.98 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa5_qualified_rate | #6 / 18 | 72.83 | ↑ higher | Source ↗official | |
| HELM Safetyanthropic_red_team | #71 / 80 | 0.957 | ↑ higher | Source ↗official | |
| HELM Safetybbq | #40 / 80 | 0.937 | ↑ higher | Source ↗official | |
| HELM Safetyharmbench | #78 / 80 | 0.397 | ↑ higher | Source ↗official | |
| HELM Safetysimple_safety_tests | #73 / 80 | 0.922 | ↑ higher | Source ↗official | |
| HELM Safetyxstest | #30 / 80 | 0.966 | ↑ higher | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #9 / 54 | 28.61 | ↑ higher | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #36 / 50 | 93.4 | ↓ lower | Source ↗official | |
| PHAREbias_resistance_diagnostic | #61 / 66 | 0.3116 | ↑ higher | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #17 / 70 | 0.8073 | ↑ higher | Source ↗official | |
| PHAREharm_resistance_diagnostic | #70 / 70 | 0.7177 | ↑ higher | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #20 / 67 | 0.6485 | ↑ higher | Source ↗official | |
| Shelleducation_jsr | #14 / 14 | 0.81 | ↓ lower | Source ↗official | |
| Shellfinance_jsr | #7 / 14 | 0.486 | ↓ lower | Source ↗official | |
| Shellmanagement_jsr | #7 / 14 | 0.596 | ↓ lower | Source ↗official | |
| Social Welfare Function Benchmarkfairness | #2 / 19 | 0.619 | ↑ higher | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #36 / 102 | 6.87 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #48 / 102 | 4.53 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #53 / 102 | 4.78 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #23 / 102 | 1.65 | ↓ lower | Source ↗official | |
| TrustLLM contemporary collapsed applicationtrustllm | #3 / 8 | 0.62 | ↑ 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.