Model profile
Gemini 3.8 Flash
Evidence summary
Safety. Gemini 3.8 Flash has an estimated Safety rank of #17; its 90% source-sensitivity interval is #11–#80. Its behavior-only rank is #20; company governance moves the combined estimate to #17. Published Safety evidence spans 15 eval lineages and 6 of 7 components. Its strongest relative result is CAIS Risk Index (agent_red_teaming, #2 of 49); its weakest is Vals AI Cheating Audit (biomystery_bench_cheating_attempt_rate_pct, #9 of 9).
Freedom. Gemini 3.8 Flash has an estimated Freedom rank of #457; its 90% source-sensitivity interval is #291–#566. Published Freedom evidence spans 2 eval lineages and 1 of 1 components. Its strongest relative result is SM-Bench (overfit, #4 of 84); its weakest is SM-Bench (adversarial, #74 of 84).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gemini-3-8-flash-medium
- OpenRouter ↗google/gemini-3.8-flash
- 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 | #83 / 345 | ↓0.519 | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #2 / 49 | ↓29.8 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #25 / 54 | ↓65.6 | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #25 / 55 | ↓51.3 | Source ↗official | |
| CAIS Risk Indexmachiavelli | #23 / 51 | ↓86.5 | Source ↗official | |
| CAIS Risk Indexmask | #39 / 57 | ↓41.4 | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #22 / 48 | ↓46.3 | Source ↗official | |
| CAIS Risk Indextextquests_harm | #49 / 54 | ↓24.4 | Source ↗official | |
| CheatBench direct cheating propensitybiology_bioinformatics_cheating_rate_pct | #5 / 8 | ↓100 | Source ↗official | |
| CheatBench direct cheating propensityboard_games_cheating_rate_pct | #6 / 8 | ↓100 | Source ↗official | |
| CheatBench direct cheating propensitycreative_writing_cheating_rate_pct | #6 / 8 | ↓100 | Source ↗official | |
| CheatBench direct cheating propensityknowledge_work_cheating_rate_pct | #1 / 8 | ↓37 | Source ↗official | |
| CheatBench direct cheating propensitymathematical_research_cheating_rate_pct | #7 / 8 | ↓98 | Source ↗official | |
| CheatBench direct cheating propensitymenial_computation_cheating_rate_pct | #4 / 8 | ↓100 | Source ↗official | |
| CheatBench direct cheating propensitymultimodal_cheating_rate_pct | #7 / 8 | ↓100 | Source ↗official | |
| CheatBench direct cheating propensitysoftware_engineering_cheating_rate_pct | #1 / 8 | ↓5 | Source ↗official | |
| CheatBench direct cheating propensitysvg_competition_cheating_rate_pct | #7 / 8 | ↓100 | Source ↗official | |
| Google Gemini 3.8 launch — Gray Swan indirect prompt injection k=15attack_success_probability_k15_pct | #2 / 15 | ↓5.5 | Source ↗official | |
| Humanity's Last Exam RMS calibration error (Scale Labs)calibrationError | #12 / 42 | ↓51 | Source ↗official | |
| Pander Scoreconversational_absolute_pander_score | #15 / 26 | ↓15.3 | Source ↗official | |
| Pander Scoreinstructional_absolute_pander_score | #20 / 26 | ↓66.04 | Source ↗official | |
| SM-Benchadversarial | #9 / 84 | ↑88.29 | Source ↗official | |
| SM-Benchambiguous_interpretation | #37 / 84 | ↑86.9 | Source ↗official | |
| SM-Benchanti_hallucination | #29 / 84 | ↑96.34 | Source ↗official | |
| SM-Bencheq_boundaries | #50 / 84 | ↑62.36 | Source ↗official | |
| SM-Benchoverfit | #4 / 84 | ↑95.08 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #53 / 123 | ↑6.83 | Source ↗official | |
| SpeciEvalland_animal_4ns | #34 / 123 | ↓4.33 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #62 / 123 | ↓4.72 | Source ↗official | |
| SpeciEvalspeciesism | #102 / 123 | ↓2.58 | Source ↗official | |
| TACbase_welfare_rate | #30 / 87 | ↑30.77 | Source ↗self run | |
| Vals AI Cheating Auditbiomystery_bench_cheating_attempt_rate_pct | #9 / 9 | ↓21.48 | Source ↗official | |
| Vals AI Cheating Auditswebench_verified_cheating_attempt_rate_pct | #2 / 6 | ↓11.6 | Source ↗official | |
| Vals AI Cheating Auditterminal_bench_cheating_shortcut_evidence_rate_pct | #12 / 14 | ↓2.622 | 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.