Model profile
Grok 4
xAIdeveloper
2025-07-09release date
#162 / 346Safety rank
#76 / 662Freedom rank
Evidence summary
Safety. Grok 4 has an estimated Safety rank of #162; its 90% source-sensitivity interval is #100–#225. Its behavior-only rank is #171; company governance moves the combined estimate to #162. Published Safety evidence spans 33 eval lineages and 7 of 7 components. Its strongest relative result is Confabulations (confabulation_rate, #3 of 52); its weakest is Concordia — MASK (safety_score, #62 of 62).
Freedom. Grok 4 has an estimated Freedom rank of #76; its 90% source-sensitivity interval is #58–#282. Published Freedom evidence spans 15 eval lineages and 1 of 1 components. Its strongest relative result is SpeechMap model completion (complete_pct, #1 of 181); its weakest is Concordia — FRT-SciKnowEval-BiologicalHarmfulQA (safety_score, #39 of 45).
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
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 |
|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #119 / 358 | ↓0.6449 | Source ↗official | |
| Adversarial Poetry — AILuminate Baseline and Poetry ASRbaseline_asr | #21 / 24 | ↓16.04 | Source ↗official | |
| Adversarial Poetry — AILuminate Baseline and Poetry ASRpoetry_asr | #9 / 24 | ↓34.4 | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #69 / 80 | ↑0.444 | Source ↗official | |
| Arena Factuality — Search Arena (factuality-only weighting)factuality_bt_rating | #26 / 30 | ↑1128.0 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #49 / 111 | ↑1440.0 | Source ↗official | |
| BrokenMathsycophancy | #5 / 9 | ↓43.4 | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #35 / 49 | ↓85.9 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #23 / 54 | ↓64.2 | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #31 / 55 | ↓56.4 | Source ↗official | |
| CAIS Risk Indexmachiavelli | #40 / 51 | ↓93.4 | Source ↗official | |
| CAIS Risk Indexmask | #40 / 57 | ↓41.7 | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #31 / 51 | ↓49.5 | Source ↗official | |
| CAIS Risk Indextextquests_harm | #48 / 54 | ↓24.3 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #39 / 104 | ↓21.78 | Source ↗official | |
| Concordia — Agentic-Misalignmentsafety_score | #53 / 54 | ↑21.83 | Source ↗official | |
| Concordia — AirBench-Deceptionsafety_score | #61 / 63 | ↑33.7 | Source ↗official | |
| Concordia — AirBench-Manipulationsafety_score | #34 / 56 | ↑82.67 | Source ↗official | |
| Concordia — AirBench-SecurityRiskssafety_score | #58 / 63 | ↑53.85 | Source ↗official | |
| Concordia — APEsafety_score | #20 / 55 | ↑41.24 | Source ↗official | |
| Concordia — CyberSecEval2-PromptInjectionsafety_score | #35 / 63 | ↑83.27 | Source ↗official | |
| Concordia — DarkBenchsafety_score | #51 / 55 | ↑39.85 | Source ↗official | |
| Concordia — Fortress-Biologicalsafety_score | #19 / 54 | ↑74.65 | Source ↗official | |
| Concordia — Fortress-Chemicalsafety_score | #24 / 54 | ↑60.64 | Source ↗official | |
| Concordia — Fortress-Privacy/Scamssafety_score | #23 / 54 | ↑65.69 | Source ↗official | |
| Concordia — FRT-AirBench-Manipulationsafety_score | #19 / 45 | ↑39.33 | Source ↗official | |
| Concordia — FRT-AirBench-SecurityRiskssafety_score | #8 / 45 | ↑47.67 | Source ↗official | |
| Concordia — FRT-SciKnowEval-BiologicalHarmfulQAsafety_score | #7 / 45 | ↑17.33 | Source ↗official | |
| Concordia — FRT-SOSBench-Chemsafety_score | #17 / 45 | ↑62.33 | Source ↗official | |
| Concordia — MASKsafety_score | #62 / 62 | ↑32.5 | Source ↗official | |
| Concordia — SciKnowEval-BiologicalHarmfulQAsafety_score | #26 / 63 | ↑85.86 | Source ↗official | |
| Concordia — SciKnowEval-ChemicalHarmfulQAsafety_score | #57 / 63 | ↑1.987 | Source ↗official | |
| Concordia — Shutdown-Resistancesafety_score | #52 / 53 | ↑58 | Source ↗official | |
| Concordia — SOSBench-Biosafety_score | #61 / 63 | ↑9.677 | Source ↗official | |
| Concordia — SOSBench-Chemsafety_score | #57 / 63 | ↑41.2 | Source ↗official | |
| Confabulationsconfabulation_rate | #3 / 52 | ↓3.96 | Source ↗official | |
| DystopiaBenchbasaglia_score | #28 / 50 | ↓65.3 | Source ↗official | |
| DystopiaBenchbaudrillard_score | #25 / 50 | ↓57.7 | Source ↗official | |
| DystopiaBenchhuxley_score | #30 / 50 | ↓73.17 | Source ↗official | |
| DystopiaBenchlaguardia_score | #15 / 50 | ↓59.07 | Source ↗official | |
| DystopiaBenchorwell_score | #25 / 50 | ↓68.63 | Source ↗official | |
| DystopiaBenchpetrov_score | #31 / 50 | ↓74.37 | Source ↗official | |
| Emergent Collusionhigh_illegality_game_rate | #13 / 13 | ↓0.75 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #206 / 270 | ↑12.4 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #213 / 270 | ↑76.83 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #250 / 270 | ↑33.89 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #163 / 268 | ↑95.45 | Source ↗official | |
| FlagEval Safety and Valuesa1_qualified_rate | #6 / 18 | ↑81.99 | Source ↗official | |
| FlagEval Safety and Valuesa2_qualified_rate | #7 / 18 | ↑80.3 | Source ↗official | |
| FlagEval Safety and Valuesa3_qualified_rate | #12 / 18 | ↑86.36 | Source ↗official | |
| FlagEval Safety and Valuesa4_qualified_rate | #10 / 18 | ↑87.98 | Source ↗official | |
| FlagEval Safety and Valuesa5_qualified_rate | #6 / 18 | ↑72.83 | Source ↗official | |
| HELM Safetyanthropic_red_team | #71 / 80 | ↑0.957 | Source ↗official | |
| HELM Safetybbq | #40 / 80 | ↑0.937 | Source ↗official | |
| HELM Safetyharmbench | #78 / 80 | ↑0.397 | Source ↗official | |
| HELM Safetysimple_safety_tests | #73 / 80 | ↑0.922 | Source ↗official | |
| HELM Safetyxstest | #30 / 80 | ↑0.966 | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #9 / 54 | ↑28.61 | Source ↗official | |
| Human Pathogen Capabilities Test (HPCT) — overall refusalrefusal_rate_pct | #27 / 69 | ↑0 | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #36 / 50 | ↓93.4 | Source ↗official | |
| PHAREbias_resistance_diagnostic | #61 / 66 | ↑0.3116 | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #17 / 70 | ↑0.8073 | Source ↗official | |
| PHAREharm_resistance_diagnostic | #70 / 70 | ↑0.7177 | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #20 / 67 | ↑0.6485 | Source ↗official | |
| RealityTest — Text AI-Identity Disclosuredisclosure_probability | #7 / 17 | ↑0.413 | Source ↗official | |
| Shelleducation_jsr | #14 / 14 | ↓0.81 | Source ↗official | |
| Shellfinance_jsr | #7 / 14 | ↓0.486 | Source ↗official | |
| Shellmanagement_jsr | #7 / 14 | ↓0.596 | Source ↗official | |
| Social Welfare Function Benchmarkfairness | #2 / 19 | ↑0.619 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #49 / 131 | ↑6.87 | Source ↗official | |
| SpeciEvalland_animal_4ns | #66 / 131 | ↓4.53 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #76 / 131 | ↓4.78 | Source ↗official | |
| SpeciEvalspeciesism | #35 / 131 | ↓1.65 | Source ↗official | |
| StereoTales Harmful Associationsbenign_significant_association_score | #16 / 23 | ↑85.53 | Source ↗official | |
| TrustLLM contemporary collapsed applicationtrustllm | #3 / 8 | ↑0.62 | 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 |
|---|---|---|---|---|
| Adversarial Poetry — AILuminate Baseline and Poetry ASRbaseline_asr | #4 / 24 | ↑16.04 | Source ↗official | |
| Adversarial Poetry — AILuminate Baseline and Poetry ASRpoetry_asr | #16 / 24 | ↑34.4 | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #12 / 80 | ↓0.444 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #32 / 54 | ↑64.2 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #66 / 104 | ↑21.78 | Source ↗official | |
| Concordia — AirBench-Deceptionsafety_score | #3 / 63 | ↓33.7 | Source ↗official | |
| Concordia — AirBench-Manipulationsafety_score | #22 / 56 | ↓82.67 | Source ↗official | |
| Concordia — AirBench-PoliticalPersuasionsafety_score | #4 / 56 | ↓24.76 | Source ↗official | |
| Concordia — AirBench-SecurityRiskssafety_score | #6 / 63 | ↓53.85 | Source ↗official | |
| Concordia — Fortress-Biologicalsafety_score | #36 / 54 | ↓74.65 | Source ↗official | |
| Concordia — Fortress-Chemicalsafety_score | #31 / 54 | ↓60.64 | Source ↗official | |
| Concordia — Fortress-Privacy/Scamssafety_score | #32 / 54 | ↓65.69 | Source ↗official | |
| Concordia — FRT-AirBench-Manipulationsafety_score | #27 / 45 | ↓39.33 | Source ↗official | |
| Concordia — FRT-AirBench-SecurityRiskssafety_score | #37 / 45 | ↓47.67 | Source ↗official | |
| Concordia — FRT-SciKnowEval-BiologicalHarmfulQAsafety_score | #39 / 45 | ↓17.33 | Source ↗official | |
| Concordia — FRT-SOSBench-Chemsafety_score | #28 / 45 | ↓62.33 | Source ↗official | |
| Concordia — SciKnowEval-BiologicalHarmfulQAsafety_score | #38 / 63 | ↓85.86 | Source ↗official | |
| Concordia — SciKnowEval-ChemicalHarmfulQAsafety_score | #7 / 63 | ↓1.987 | Source ↗official | |
| Concordia — SOSBench-Biosafety_score | #3 / 63 | ↓9.677 | Source ↗official | |
| Concordia — SOSBench-Chemsafety_score | #7 / 63 | ↓41.2 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #58 / 270 | ↓76.83 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #20 / 270 | ↓33.89 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #101 / 268 | ↓95.45 | Source ↗official | |
| HELM Safetyanthropic_red_team | #10 / 80 | ↓0.957 | Source ↗official | |
| HELM Safetyharmbench | #3 / 80 | ↓0.397 | Source ↗official | |
| HELM Safetysimple_safety_tests | #8 / 80 | ↓0.922 | Source ↗official | |
| HELM Safetyxstest | #30 / 80 | ↑0.966 | Source ↗official | |
| Human Pathogen Capabilities Test (HPCT) — overall refusalrefusal_rate_pct | #1 / 69 | ↓0 | Source ↗official | |
| PHAREharm_resistance_diagnostic | #1 / 70 | ↓0.7177 | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #48 / 67 | ↓0.6485 | Source ↗official | |
| Shelleducation_jsr | #1 / 14 | ↑0.81 | Source ↗official | |
| Shellfinance_jsr | #8 / 14 | ↑0.486 | Source ↗official | |
| Shellmanagement_jsr | #8 / 14 | ↑0.596 | Source ↗official | |
| SpeechMap model completioncomplete_pct | #1 / 181 | ↑98.3 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_adherence_score | #25 / 156 | ↑6 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_direct_score | #4 / 156 | ↑6 | 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.