Model profile
Gemini 1.5 Flash
Googledeveloper
2024-05-14release date
#93 / 267overall rank
15eval lineages
Evidence summary
Gemini 1.5 Flash has an estimated overall rank of #93; its 90% source-sensitivity interval is #36–#167. Its behavior-only rank is #97; company governance moves the combined estimate to #93. Published evidence spans 15 evals and 6 of 7 behavior components. Its strongest relative result is HELM Safety (anthropic_red_team, #2 of 80); its weakest is Enkrypt AI Safety Leaderboard (toxicity_attack_non_success_rate, #214 of 258).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gemini-1-5-flash
- Official model documentation ↗Family-level model document · google · 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 | Better | Source |
|---|---|---|---|---|---|
| Adversarial Robustnessscore | #4 / 8 | 14 | ↓ lower | Source ↗official | |
| Agent-SafetyBenchcompromise_availability | #9 / 16 | 30 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchharmful_vulnerable_code | #4 / 16 | 48.4 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchleak_sensitive_information | #5 / 16 | 39.2 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchphysical_harm | #6 / 16 | 38.8 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchproduce_unsafe_information | #11 / 16 | 82.4 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchproperty_loss | #7 / 16 | 41.6 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchspread_unsafe_information | #4 / 16 | 20.8 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchviolate_law_ethics | #6 / 16 | 32 | ↑ higher | Source ↗official | |
| AgentDojotargeted_attack_success_rate | #4 / 15 | 0.0787 | ↓ lower | Source ↗official | |
| AgentDojoutility_under_attack | #11 / 15 | 0.333 | ↑ higher | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #31 / 80 | 0.7325 | ↑ higher | Source ↗official | |
| AnimalHarmBenchscore | #3 / 10 | 0.05 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #199 / 260 | 12.14 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #162 / 260 | 86.67 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #170 / 260 | 55.56 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #214 / 258 | 90.59 | ↑ higher | Source ↗official | |
| FORTRESSaverage_risk_score | #35 / 49 | 50.61 | ↓ lower | Source ↗official | |
| FORTRESSover_refusal_score | #22 / 46 | 4.45 | ↓ lower | Source ↗official | |
| HELM Safetyanthropic_red_team | #2 / 80 | 0.999 | ↑ higher | Source ↗official | |
| HELM Safetybbq | #32 / 80 | 0.947 | ↑ higher | Source ↗official | |
| HELM Safetyharmbench | #35 / 80 | 0.8 | ↑ higher | Source ↗official | |
| HELM Safetysimple_safety_tests | #58 / 80 | 0.97 | ↑ higher | Source ↗official | |
| HELM Safetyxstest | #66 / 80 | 0.921 | ↑ higher | Source ↗official | |
| Large-scale Moral Machine experiment on LLMshuman_choice_distance | #23 / 39 | 1.116 | ↓ lower | Source ↗official | |
| OR-Benchover_refusal_rate | #17 / 25 | 84.3 | ↓ lower | Source ↗official | |
| OR-Benchtoxic_acceptance_rate | #8 / 25 | 1.2 | ↓ lower | Source ↗official | |
| SORRY-Benchavg | #5 / 51 | 0.08 | ↓ lower | Source ↗official |