Model profile
Grok 4.5
Evidence summary
Grok 4.5 has an estimated overall rank of #58; its 90% source-sensitivity interval is #32–#198. Its behavior-only rank is #50; company governance moves the combined estimate to #58. Published evidence spans 15 evals and 6 of 7 behavior components. Its strongest relative result is SpeciEval (belief_animal_sentience, #1 of 113); its weakest is Gray Swan indirect prompt injection (15 attempts) (attack_success_probability_k15_pct, #13 of 13).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗grok-4-5
- OpenRouter ↗x-ai/grok-4.5
- 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 | #68 / 328 | ↓0.5415 | Source ↗official | |
| Arena Factuality — Search Arena (factuality-only weighting)factuality_bt_rating | #5 / 30 | ↑1227.0 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #10 / 112 | ↑1468.0 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #20 / 106 | ↑0.54 | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #27 / 44 | ↓83.5 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #34 / 49 | ↓79.2 | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #6 / 50 | ↓30.8 | Source ↗official | |
| CAIS Risk Indexmachiavelli | #29 / 46 | ↓90.9 | Source ↗official | |
| CAIS Risk Indexmask | #32 / 52 | ↓27.9 | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #22 / 33 | ↓50.2 | Source ↗official | |
| CAIS Risk Indextextquests_harm | #5 / 49 | ↓11.2 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #40 / 104 | ↓22.98 | Source ↗official | |
| Gray Swan indirect prompt injection (15 attempts)attack_success_probability_k15_pct | #13 / 13 | ↓60.8 | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #30 / 50 | ↓90.9 | Source ↗official | |
| Olam Social Poker — Social Lie Ratesocial_lie_rate_per_10000_turns | #10 / 19 | ↓9 | Source ↗official | |
| SM-Benchadversarial | #21 / 79 | ↑85.37 | Source ↗official | |
| SM-Benchambiguous_interpretation | #8 / 79 | ↑91.96 | Source ↗official | |
| SM-Benchanti_hallucination | #18 / 79 | ↑97.91 | Source ↗official | |
| SM-Bencheq_boundaries | #20 / 79 | ↑69.1 | Source ↗official | |
| SM-Benchoverfit | #8 / 79 | ↑92.35 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #1 / 113 | ↑7 | Source ↗official | |
| SpeciEvalland_animal_4ns | #109 / 113 | ↓5.15 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #103 / 113 | ↓5.15 | Source ↗official | |
| SpeciEvalspeciesism | #111 / 113 | ↓3.52 | 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.