Model profile
Grok 4 Fast
Evidence summary
Safety. Grok 4 Fast has an estimated Safety rank of #134; its 90% source-sensitivity interval is #64–#239. Its behavior-only rank is #130; company governance moves the combined estimate to #134. Published Safety evidence spans 18 eval lineages and 6 of 7 components. Its strongest relative result is PHARE (bias_resistance_diagnostic, #2 of 66); its weakest is CAIS Risk Index (mask, #55 of 57).
Freedom. Grok 4 Fast has an estimated Freedom rank of #376; its 90% source-sensitivity interval is #164–#533. Published Freedom evidence spans 6 eval lineages and 1 of 1 components. Its strongest relative result is PHARE (harm_resistance_diagnostic, #7 of 70); its weakest is Human Pathogen Capabilities Test (HPCT) — overall refusal (refusal_rate_pct, #53 of 69).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗grok-4-fast
- OpenRouter ↗x-ai/grok-4-fast
- 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 | #121 / 345 | ↓0.6827 | Source ↗official | |
| Adversarial Poetry — AILuminate Baseline and Poetry ASRbaseline_asr | #14 / 24 | ↓7.84 | Source ↗official | |
| Adversarial Poetry — AILuminate Baseline and Poetry ASRpoetry_asr | #10 / 24 | ↓35.58 | Source ↗official | |
| AgentDrive Safety Compliancescr | #20 / 48 | ↑88.75 | Source ↗official | |
| ANIMAscore | #16 / 22 | ↑0.6164 | Source ↗official | |
| Arena Factuality — Search Arena (factuality-only weighting)factuality_bt_rating | #28 / 30 | ↑1112.0 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #67 / 111 | ↑1430.0 | Source ↗official | |
| BrokenMathsycophancy | #4 / 9 | ↓40 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #22 / 54 | ↓64.1 | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #41 / 55 | ↓68.7 | Source ↗official | |
| CAIS Risk Indexmachiavelli | #46 / 51 | ↓96.2 | Source ↗official | |
| CAIS Risk Indexmask | #55 / 57 | ↓70 | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #2 / 48 | ↓30.8 | Source ↗official | |
| CAIS Risk Indextextquests_harm | #30 / 54 | ↓19.8 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #43 / 104 | ↓30.03 | Source ↗official | |
| Human Pathogen Capabilities Test (HPCT) — overall refusalrefusal_rate_pct | #17 / 69 | ↑19.5 | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #43 / 50 | ↓96.2 | Source ↗official | |
| NESSiE Necessary Safety Benchmarksh_score | #5 / 16 | ↑92.02 | Source ↗official | |
| PHAREbias_resistance_diagnostic | #2 / 66 | ↑0.8026 | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #62 / 70 | ↑0.634 | Source ↗official | |
| PHAREharm_resistance_diagnostic | #64 / 70 | ↑0.8134 | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #49 / 67 | ↑0.412 | Source ↗official | |
| UAVBench safety-critical decision recognitionethical_safety_critical_accuracy | #14 / 27 | ↑0.695 | Source ↗official | |
| Vectara HHEM Factual Consistencyfactual_consistency_rate | #90 / 94 | ↑80.05 | 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 | #11 / 24 | ↑7.84 | Source ↗official | |
| Adversarial Poetry — AILuminate Baseline and Poetry ASRpoetry_asr | #15 / 24 | ↑35.58 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #33 / 54 | ↑64.1 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #62 / 104 | ↑30.03 | Source ↗official | |
| Human Pathogen Capabilities Test (HPCT) — overall refusalrefusal_rate_pct | #53 / 69 | ↓19.5 | Source ↗official | |
| PHAREharm_resistance_diagnostic | #7 / 70 | ↓0.8134 | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #19 / 67 | ↓0.412 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_adherence_score | #35 / 156 | ↑5.25 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_direct_score | #66 / 156 | ↑3.5 | 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.