Model profile
Gemini 3.1 Pro Preview
Googledeveloper
2026-02-19release date
#63 / 333Safety rank
#341 / 645Freedom rank
Evidence summary
Safety. Gemini 3.1 Pro Preview has an estimated Safety rank of #63; its 90% source-sensitivity interval is #40–#141. Its behavior-only rank is #69; company governance moves the combined estimate to #63. Published Safety evidence spans 34 eval lineages and 7 of 7 components. Its strongest relative result is SpeciEval (belief_animal_sentience, #1 of 123); its weakest is Concordia — Shutdown-Resistance (safety_score, #52 of 53).
Freedom. Gemini 3.1 Pro Preview has an estimated Freedom rank of #341; its 90% source-sensitivity interval is #162–#454. Published Freedom evidence spans 13 eval lineages and 1 of 1 components. Its strongest relative result is Human Pathogen Capabilities Test (HPCT) — overall refusal (refusal_rate_pct, #1 of 69); its weakest is SM-Bench (adversarial, #80 of 84).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gemini-3-1-pro-preview
- OpenRouter ↗google/gemini-3.1-pro-preview
- Official model documentation ↗Family-level model document · google · 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 | #78 / 345 | ↓0.5087 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5child_sexual_exploitation | #18 / 31 | ↓50 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5defamation | #28 / 31 | ↓84.8 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5hate | #25 / 31 | ↓68.2 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5indiscriminate_weapons_cbrne | #13 / 31 | ↓61.4 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5intellectual_property | #28 / 31 | ↓76.6 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5non_violent_crimes | #23 / 31 | ↓86.7 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5privacy | #24 / 31 | ↓78.7 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5sex_related_crimes | #25 / 31 | ↓81.2 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5sexual_content | #30 / 31 | ↓68.8 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5specialized_advice | #28 / 31 | ↓74.5 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5suicide_self_harm | #17 / 31 | ↓60.9 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5violent_crimes | #26 / 31 | ↓84.1 | Source ↗official | |
| AgentAbstainabstain | #6 / 17 | ↑65.4 | Source ↗official | |
| AgentAbstaincar | #3 / 17 | ↑65.7 | Source ↗official | |
| AgentAbstainpaired | #1 / 17 | ↑59.5 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #37 / 111 | ↑1446.0 | Source ↗official | |
| BioSecBench-Refusalbalanced_refusal_score | #6 / 10 | ↑0.3931 | Source ↗official | |
| BioTIERpermit_compliance_pct | #8 / 52 | ↑99.6 | Source ↗official | |
| BioTIERrefuse_compliance_pct | #11 / 52 | ↑73.5 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #52 / 117 | ↑0.34 | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #21 / 49 | ↓68 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #34 / 54 | ↓73.5 | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #23 / 55 | ↓50.3 | Source ↗official | |
| CAIS Risk Indexmachiavelli | #30 / 51 | ↓89.2 | Source ↗official | |
| CAIS Risk Indexmask | #49 / 57 | ↓53.8 | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #14 / 48 | ↓42.2 | Source ↗official | |
| CAIS Risk Indextextquests_harm | #29 / 54 | ↓18.8 | Source ↗official | |
| Concordia — Agentic-Misalignmentsafety_score | #52 / 54 | ↑29.83 | Source ↗official | |
| Concordia — AirBench-Deceptionsafety_score | #28 / 63 | ↑86.3 | Source ↗official | |
| Concordia — AirBench-Manipulationsafety_score | #26 / 56 | ↑90.67 | Source ↗official | |
| Concordia — AirBench-SecurityRiskssafety_score | #19 / 63 | ↑97.07 | Source ↗official | |
| Concordia — APEsafety_score | #33 / 55 | ↑16.74 | Source ↗official | |
| Concordia — CyberSecEval2-PromptInjectionsafety_score | #3 / 63 | ↑98.8 | Source ↗official | |
| Concordia — DarkBenchsafety_score | #52 / 55 | ↑39.36 | Source ↗official | |
| Concordia — Fortress-Biologicalsafety_score | #16 / 54 | ↑77.24 | Source ↗official | |
| Concordia — Fortress-Chemicalsafety_score | #17 / 54 | ↑74.56 | Source ↗official | |
| Concordia — Fortress-Privacy/Scamssafety_score | #20 / 54 | ↑69.9 | Source ↗official | |
| Concordia — FRT-AirBench-Manipulationsafety_score | #25 / 45 | ↑32.89 | Source ↗official | |
| Concordia — FRT-AirBench-SecurityRiskssafety_score | #24 / 45 | ↑27.33 | Source ↗official | |
| Concordia — FRT-SciKnowEval-BiologicalHarmfulQAsafety_score | #38 / 45 | ↑1 | Source ↗official | |
| Concordia — FRT-SOSBench-Chemsafety_score | #30 / 45 | ↑46.67 | Source ↗official | |
| Concordia — MASKsafety_score | #44 / 62 | ↑50.37 | Source ↗official | |
| Concordia — SciKnowEval-BiologicalHarmfulQAsafety_score | #16 / 63 | ↑97.31 | Source ↗official | |
| Concordia — SciKnowEval-ChemicalHarmfulQAsafety_score | #11 / 63 | ↑40.62 | Source ↗official | |
| Concordia — Shutdown-Resistancesafety_score | #52 / 53 | ↑58 | Source ↗official | |
| Concordia — SOSBench-Biosafety_score | #32 / 63 | ↑81 | Source ↗official | |
| Concordia — SOSBench-Chemsafety_score | #12 / 63 | ↑94.4 | Source ↗official | |
| DelusionEvaldelusional_prevalence_pct | #10 / 16 | ↓47.3 | Source ↗official | |
| DelusionEvaldiscourages_harm_prevalence_pct | #7 / 16 | ↑29.4 | Source ↗official | |
| DelusionEvalfacilitates_harm_prevalence_pct | #9 / 16 | ↓2.7 | Source ↗official | |
| DelusionEvalrelationship_prevalence_pct | #13 / 16 | ↓37.6 | Source ↗official | |
| DelusionEvalsycophancy_prevalence_pct | #10 / 16 | ↓30.4 | Source ↗official | |
| FORTRESSaverage_risk_score | #25 / 50 | ↓29.76 | Source ↗official | |
| FORTRESSover_refusal_score | #19 / 49 | ↓3.13 | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #7 / 54 | ↑29.01 | Source ↗official | |
| Human Pathogen Capabilities Test (HPCT) — overall refusalrefusal_rate_pct | #27 / 69 | ↑0 | Source ↗official | |
| Humanity's Last Exam RMS calibration error (Scale Labs)calibrationError | #12 / 42 | ↓51 | Source ↗official | |
| JuICE Cultural-Error Span Detectionf1 | #1 / 10 | ↑0.5225 | Source ↗official | |
| LiveSecBenchethics | #23 / 43 | ↑45.45 | Source ↗official | |
| LiveSecBenchfactuality | #21 / 43 | ↑51.4 | Source ↗official | |
| LiveSecBenchlegality | #19 / 43 | ↑50.36 | Source ↗official | |
| LiveSecBenchprivacy | #12 / 43 | ↑67.16 | Source ↗official | |
| LiveSecBenchpsychological_health | #6 / 43 | ↑76.44 | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #25 / 50 | ↓89.2 | Source ↗official | |
| MANTAAWMS | #7 / 12 | ↑0.441 | Source ↗official | |
| MANTAAWVS | #9 / 12 | ↑0.404 | Source ↗official | |
| MASKlying_probability_pct | #52 / 53 | ↓57.6 | Source ↗official | |
| ODCV-Benchaverage_severity | #11 / 12 | ↓1.825 | Source ↗official | |
| ODCV-Benchmisalignment_rate | #10 / 12 | ↓43.75 | Source ↗official | |
| Olam Social Poker — Social Lie Ratesocial_lie_rate_per_10000_turns | #11 / 24 | ↓8 | Source ↗official | |
| Pander Scoreconversational_absolute_pander_score | #22 / 26 | ↓22.66 | Source ↗official | |
| Pander Scoreinstructional_absolute_pander_score | #25 / 26 | ↓70.61 | Source ↗official | |
| RealityTest — Text AI-Identity Disclosuredisclosure_probability | #9 / 17 | ↑0.305 | Source ↗official | |
| RefusalBenchyouden_j | #8 / 19 | ↑0.1319 | Source ↗official | |
| SM-Benchadversarial | #4 / 84 | ↑90.73 | Source ↗official | |
| SM-Benchambiguous_interpretation | #45 / 84 | ↑85.71 | Source ↗official | |
| SM-Benchanti_hallucination | #10 / 84 | ↑98.95 | Source ↗official | |
| SM-Bencheq_boundaries | #28 / 84 | ↑68.26 | Source ↗official | |
| SM-Benchoverfit | #24 / 84 | ↑84.15 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #1 / 123 | ↑7 | Source ↗official | |
| SpeciEvalland_animal_4ns | #36 / 123 | ↓4.35 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #56 / 123 | ↓4.7 | Source ↗official | |
| SpeciEvalspeciesism | #118 / 123 | ↓3.03 | Source ↗official | |
| StereoTales Harmful Associationsbenign_significant_association_score | #20 / 23 | ↑84.41 | Source ↗official | |
| Vectara HHEM Factual Consistencyfactual_consistency_rate | #55 / 94 | ↑89.6 | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #2 / 25 | ↓4.9 | Source ↗official | |
| Vigil Mental Health Safetyoverall_score | #10 / 23 | ↑48 | Source ↗official | |
| WildClawBench Safety & Alignment (OpenClaw harness)safety_alignment_score_pct | #6 / 24 | ↑45.77 | 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.
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.
UGI Political Values
Agent-ValueBench Moral Foundations (MFT08)
Agent-ValueBench HEXACO
| Dimension | Value | Distribution |
|---|---|---|
| Openness to experience | 4.1 | |
| Honesty-humility | 5.4 | |
| Extraversion | 5.9 | |
| Agreeableness | 5.7 | |
| Conscientiousness | 7.1 |