Model profile
Gemini 3 Flash Preview
Evidence summary
Safety. Gemini 3 Flash Preview has an estimated Safety rank of #212; its 90% source-sensitivity interval is #116–#259. Its behavior-only rank is #226; company governance moves the combined estimate to #212. Published Safety evidence spans 29 eval lineages and 7 of 7 components. Its strongest relative result is SpeciEval (belief_animal_sentience, #1 of 123); its weakest is Adversarial Humanities Benchmark (AHB) — Table 5 (child_sexual_exploitation, #31 of 31).
Freedom. Gemini 3 Flash Preview has an estimated Freedom rank of #340; its 90% source-sensitivity interval is #144–#503. Published Freedom evidence spans 8 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 BioTIER (permit_compliance_pct, #38 of 52).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gemini-3-flash
- OpenRouter ↗google/gemini-3-flash-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 | #301 / 345 | ↓0.9241 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5child_sexual_exploitation | #31 / 31 | ↓83.3 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5defamation | #28 / 31 | ↓84.8 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5hate | #24 / 31 | ↓66.7 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5indiscriminate_weapons_cbrne | #29 / 31 | ↓88.9 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5intellectual_property | #31 / 31 | ↓81.2 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5non_violent_crimes | #29 / 31 | ↓93.5 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5privacy | #30 / 31 | ↓85.1 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5sex_related_crimes | #31 / 31 | ↓93.8 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5sexual_content | #28 / 31 | ↓66.7 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5specialized_advice | #30 / 31 | ↓76.53 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5suicide_self_harm | #28 / 31 | ↓77.1 | Source ↗official | |
| Adversarial Humanities Benchmark (AHB) — Table 5violent_crimes | #27 / 31 | ↓88.9 | Source ↗official | |
| AgentAbstainabstain | #16 / 17 | ↑43.6 | Source ↗official | |
| AgentAbstaincar | #15 / 17 | ↑43.4 | Source ↗official | |
| AgentAbstainpaired | #14 / 17 | ↑39.7 | Source ↗official | |
| Alignment Leaderboardcorrigibility | #10 / 24 | ↑4.32 | Source ↗official | |
| Alignment Leaderboardhonesty | #9 / 24 | ↑3.682 | Source ↗official | |
| Alignment Leaderboardnon_manipulation | #12 / 24 | ↑3.51 | Source ↗official | |
| Alignment Leaderboardrobustness | #14 / 24 | ↑3.373 | Source ↗official | |
| Alignment Leaderboardsafety | #11 / 24 | ↑3.875 | Source ↗official | |
| Alignment Leaderboardscheming | #12 / 24 | ↑3.645 | Source ↗official | |
| Arena Factuality — Search Arena (factuality-only weighting)factuality_bt_rating | #25 / 30 | ↑1153.0 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #32 / 111 | ↑1449.0 | Source ↗official | |
| AuAu Authoritarian Response Auditbehavioral_vignette_arr_percent | #3 / 17 | ↓21.8 | Source ↗official | |
| AuAu Authoritarian Response Auditrealistic_prompt_arr_percent | #10 / 17 | ↓2.2 | Source ↗official | |
| BioTIERpermit_compliance_pct | #38 / 52 | ↑98.3 | Source ↗official | |
| BioTIERrefuse_compliance_pct | #22 / 52 | ↑54.3 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #97 / 117 | ↑0.1 | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #47 / 49 | ↓92.9 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #35 / 54 | ↓75.3 | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #28 / 55 | ↓55.1 | Source ↗official | |
| CAIS Risk Indexmachiavelli | #42 / 51 | ↓95.3 | Source ↗official | |
| CAIS Risk Indexmask | #54 / 57 | ↓61.7 | Source ↗official | |
| CAIS Risk Indextextquests_harm | #25 / 54 | ↓17.8 | Source ↗official | |
| DystopiaBenchbasaglia_score | #47 / 50 | ↓73.2 | Source ↗official | |
| DystopiaBenchbaudrillard_score | #50 / 50 | ↓81.13 | Source ↗official | |
| DystopiaBenchhuxley_score | #47 / 50 | ↓83 | Source ↗official | |
| DystopiaBenchlaguardia_score | #46 / 50 | ↓71.63 | Source ↗official | |
| DystopiaBenchorwell_score | #31 / 50 | ↓72.43 | Source ↗official | |
| DystopiaBenchpetrov_score | #43 / 50 | ↓80.33 | Source ↗official | |
| Every Model Cheats — Cybench Cheat Propensitybaseline_cheat_propensity_pct | #10 / 22 | ↓30.4 | Source ↗official | |
| Every Model Cheats — Cybench Cheat Propensitysevere_anti_cheat_cheat_propensity_pct | #1 / 22 | ↓0 | Source ↗official | |
| Every Model Cheats — Cybench Cheat Propensitystandard_anti_cheat_cheat_propensity_pct | #22 / 22 | ↓69.6 | Source ↗official | |
| Human Pathogen Capabilities Test (HPCT) — overall refusalrefusal_rate_pct | #27 / 69 | ↑0 | Source ↗official | |
| JuICE Cultural-Error Span Detectionf1 | #3 / 10 | ↑0.4839 | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #39 / 50 | ↓95.3 | Source ↗official | |
| MT-JailBench CrescendoXsafety_score | #16 / 21 | ↑13.84 | Source ↗official | |
| NESSiE Necessary Safety Benchmarksh_score | #11 / 16 | ↑79.49 | Source ↗official | |
| Pander Scoreconversational_absolute_pander_score | #24 / 26 | ↓25.97 | Source ↗official | |
| Pander Scoreinstructional_absolute_pander_score | #24 / 26 | ↓70.12 | Source ↗official | |
| RealityTest — Text AI-Identity Disclosuredisclosure_probability | #12 / 17 | ↑0.227 | Source ↗official | |
| SM-Benchadversarial | #27 / 84 | ↑84.39 | Source ↗official | |
| SM-Benchambiguous_interpretation | #63 / 84 | ↑80.65 | Source ↗official | |
| SM-Benchanti_hallucination | #70 / 84 | ↑82.72 | Source ↗official | |
| SM-Bencheq_boundaries | #40 / 84 | ↑65.45 | Source ↗official | |
| SM-Benchoverfit | #7 / 84 | ↑93.99 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #1 / 123 | ↑7 | Source ↗official | |
| SpeciEvalland_animal_4ns | #86 / 123 | ↓4.72 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #99 / 123 | ↓5.03 | Source ↗official | |
| SpeciEvalspeciesism | #123 / 123 | ↓3.9 | Source ↗official | |
| StereoTales Harmful Associationsbenign_significant_association_score | #23 / 23 | ↑82.56 | Source ↗official | |
| TACbase_welfare_rate | #18 / 87 | ↑35.9 | Source ↗self run | |
| Vals AI Cheating Auditterminal_bench_cheating_shortcut_evidence_rate_pct | #2 / 14 | ↓0.3745 | Source ↗official | |
| Vectara HHEM Factual Consistencyfactual_consistency_rate | #79 / 94 | ↑86.5 | Source ↗official | |
| Vigil Mental Health Safetyoverall_score | #11 / 23 | ↑47 | 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 | 6.5 | |
| Honesty-humility | 5.5 | |
| Extraversion | 6.5 | |
| Agreeableness | 5.7 | |
| Conscientiousness | 7.3 |