Google · model
Gemini 3 Flash benchmark scores
As of 2026-11-21, Gemini 3 Flash (Google) has measured scores on 31 benchmarks and estimated scores on 119 more.
| Benchmark | Score | Source |
|---|---|---|
| AA-GPQA Diamond | 81.2% | measured |
| AA-HLE | 15.0% | measured |
| AA-SciCode | 48.5% | estimated ± 3.8 pp, high confidence |
| AIME26 | 95.9% | estimated ± 1.2 pp, medium confidence |
| Artificial Analysis Intelligence Index | 17.9% | measured |
| CritPt | 1.4% | measured |
| GDPval-AA | 21.4% | estimated ± 6.2 pp, medium confidence |
| GPQA Diamond | 89.7% | estimated ± 1.8 pp, high confidence |
| HLE | 20.4% | estimated ± 8.5 pp, medium confidence |
| IFBench | 57.7% | estimated ± 6.6 pp, medium confidence |
| AA-LCR | 55.3% | measured |
| LiveCodeBench v6 | 89.0% | estimated ± 2.2 pp, medium confidence |
| AA-Omniscience Accuracy | 45.8% | measured |
| SciCode | 47.9% | estimated ± 2.9 pp, medium confidence |
| Terminal-Bench 2.1 | 69.9% | estimated ± 6.7 pp, medium confidence |
| BrowseComp | 64.0% | estimated ± 6.2 pp, medium confidence |
| IFEval | 91.7% | estimated ± 2.8 pp, high confidence |
| LongBench v2 | 53.1% | estimated ± 5.5 pp, medium confidence |
| VITA-Bench | 28.6% | estimated ± 10.7 pp, low confidence |
| τ²-bench results | 43.3% | measured |
| AA Coding Index | 59.3% | estimated ± 5.6 pp, medium confidence |
| AA-MMMU-Pro | 78.6% | measured |
| APEX-Agents-AA | 15.5% | estimated ± 3.6 pp, medium confidence |
| BioMysteryBench (human-difficult) | 20.9% | estimated ± 6.8 pp, low confidence |
| DeepSearchQA | 73.8% | estimated ± 8.2 pp, low confidence |
| SWE-bench Verified | 79.3% | estimated ± 4.4 pp, high confidence |
| BFCL v4 | 38.7% | estimated ± 8.5 pp, low confidence |
| CyberGym | 55.3% | estimated ± 9.3 pp, medium confidence |
| JobBench | 11.4% | measured |
| SWE-bench Pro | 54.6% | estimated ± 2.9 pp, medium confidence |
| WideResearch | 72.8% | estimated ± 1.2 pp, medium confidence |
| AA Agentic Index | 21.1% | estimated ± 6.8 pp, medium confidence |
| MMLU-Pro | 86.9% | estimated ± 4.8 pp, high confidence |
| AA-IFBench | 55.1% | measured |
| FailureSensorIQ | 67.5% | estimated ± 1.7 pp, high confidence |
| GPQA | 89.5% | estimated ± 4.5 pp, high confidence |
| MMLU | 90.5% | estimated ± 0.9 pp, medium confidence |
| Vals GPQA Diamond | 87.9% | measured |
| Vals LiveCodeBench | 85.6% | measured |
| Vals MMLU-Pro | 88.6% | measured |
| MMMLU | 86.8% | estimated ± 0.7 pp, medium confidence |
| MMMU | 85.6% | estimated ± 2.7 pp, high confidence |
| Chess Puzzles | 40.0% | measured |
| AA MMLU-Pro | 88.2% | estimated ± 0.7 pp, high confidence |
| Gert Labs | 56.6% | measured |
| AA EnterpriseOps-Gym | 37.6% | estimated ± 1.2 pp, low confidence |
| AA Terminal-Bench 2.1 | 80.0% | estimated ± 2.5 pp, medium confidence |
| ApprenticeBench | 0.0% | estimated ± 3.1 pp, low confidence |
| ARC-AGI-1 | 83.4% | estimated ± 6.5 pp, medium confidence |
| ARC-AGI-2 | 50.3% | estimated ± 13.8 pp, low confidence |
| CursorBench 3.1 | 49.0% | estimated ± 5.8 pp, low confidence |
| CursorBench 3.2 | 55.5% | estimated ± 5.4 pp, low confidence |
| EBR-bench | 11.6% | estimated ± 8.1 pp, medium confidence |
| Terminal-Bench 3.0 | 3.1% | estimated ± 6.9 pp, low confidence |
| FrontierCode 1.1 Main | 30.6% | estimated ± 5.2 pp, medium confidence |
| FrontierMath v2 (Tier 4) | 17.1% | measured |
| FrontierMath v2 (Tiers 1-3) | 51.2% | measured |
| FrontierSWE v2 | 8.0% | estimated ± 9.5 pp, low confidence |
| Furniture Assembly | 32.1% | estimated ± 7.1 pp, medium confidence |
| MaxxBench | 0.0% | estimated ± 5.6 pp, low confidence |
| MazeBench | 0.0% | estimated ± 5.1 pp, low confidence |
| Mystery Game Puzzles | 20.0% | measured |
| NYT Connections | 83.1% | measured |
| OfficeQA Pro | 50.8% | estimated ± 6.8 pp, medium confidence |
| OSWorld-Verified | 58.7% | estimated ± 3.8 pp, low confidence |
| ParseBench-1 | 74.1% | measured |
| Terminal-Bench 4.1 | 9.2% | estimated ± 14.9 pp, low confidence |
| Terminal-Bench Hard | 45.8% | estimated ± 3.9 pp, high confidence |
| Vals SWE-bench | 75.0% | measured |
| Terminal-Bench 2.1 (Vals AI) | 53.9% | measured |
| VulcanBench v3 | 84.0% | estimated ± 5.6 pp, low confidence |
| HealthBench Professional (length-adjusted, Artificial Analysis) | 51.3% | estimated ± 4.3 pp, low confidence |
| HealthBench Professional (raw, Artificial Analysis) | 53.4% | estimated ± 1.1 pp, low confidence |
| AA ITBench | 39.2% | estimated ± 2.8 pp, high confidence |
| Chartography (tools) | 82.5% | estimated ± 4.1 pp, low confidence |
| CursorBench 4.0 | 30.8% | estimated ± 3.4 pp, medium confidence |
| DeepSWE | 56.3% | estimated ± 7.2 pp, low confidence |
| FrontierPhysics | 0.0% | estimated ± 4.6 pp, medium confidence |
| HLE w/o tools | 45.3% | estimated ± 4.9 pp, high confidence |
| OSWorld 2.0 | 5.7% | estimated ± 5.9 pp, low confidence |
| PostTrainBench v1.1 | 19.8% | estimated ± 3.7 pp, medium confidence |
| SpireClimb | 5.8% | estimated ± 7.2 pp, medium confidence |
| StealthMark | 30.5% | estimated ± 3.5 pp, medium confidence |
| SWE-bench Multimodal | 57.0% | estimated ± 5.6 pp, low confidence |
| GDP.pdf | 16.4% | estimated ± 4.1 pp, low confidence |
| AA Terminal-Bench 4.0 | 0.0% | estimated ± 13.7 pp, low confidence |
| ARC-AGI-3 | 3.6% | estimated ± 8.6 pp, low confidence |
| ARC-AGI-3 Provider Adapter | 2.7% | estimated ± 7.3 pp, low confidence |
| SkillsBench official leaderboard (BenchFlow) | 54.6% | measured |
| BinaryAudit | 36.8% | measured |
| CompileBench | 84.4% | estimated ± 10.0 pp, low confidence |
| OTelBench | 18.8% | measured |
| Vibe Code Bench | 20.2% | measured |
| FACTS Multimodal | 49.7% | measured |
| ParseBench-2 | 77.7% | measured |
| CharXiv | 88.9% | estimated ± 5.6 pp, low confidence |
| CharXiv w/o tools | 87.4% | estimated ± 1.8 pp, low confidence |
| ExploitBench | 31.5% | estimated ± 1.9 pp, low confidence |
| AI2D_TEST | 94.0% | estimated ± 2.0 pp, medium confidence |
| C-Eval | 90.9% | estimated ± 1.0 pp, medium confidence |
| Claw-Eval | 49.2% | measured |
| CountBench | 97.3% | estimated ± 2.9 pp, medium confidence |
| DynaMath | 85.5% | estimated ± 3.0 pp, medium confidence |
| ERQA | 68.4% | estimated ± 3.1 pp, high confidence |
| HMMT Feb 2026 | 89.9% | estimated ± 1.5 pp, medium confidence |
| HMMT Nov 2025 | 94.5% | estimated ± 1.5 pp, medium confidence |
| MathVision | 89.3% | estimated ± 3.1 pp, high confidence |
| MCP Atlas | 29.2% | estimated ± 8.6 pp, medium confidence |
| MCP-Tasks | 54.8% | estimated ± 6.4 pp, low confidence |
| MMAnswerBench | 86.9% | estimated ± 2.4 pp, medium confidence |
| MMLU-Redux | 96.7% | estimated ± 0.8 pp, low confidence |
| MMMU-Pro | 79.4% | estimated ± 1.5 pp, high confidence |
| NL2Repo | 42.5% | estimated ± 4.7 pp, high confidence |
| PostTrainBench v1 | 19.0% | estimated ± 3.1 pp, medium confidence |
| RealWorldQA | 85.3% | estimated ± 2.9 pp, medium confidence |
| ScreenSpot Pro | 92.2% | estimated ± 10.1 pp, low confidence |
| SuperGPQA | 91.9% | estimated ± 9.3 pp, medium confidence |
| Terminal-Bench 2.0 | 61.7% | estimated ± 5.8 pp, medium confidence |
| Toolathlon | 46.1% | estimated ± 7.2 pp, medium confidence |
| V* | 100.0% | estimated ± 7.5 pp, medium confidence |
| Video-MME (w/o subtitle) | 85.6% | estimated ± 1.7 pp, medium confidence |
| Video-MME (with subtitle) | 87.4% | estimated ± 3.7 pp, medium confidence |
| AA LiveCodeBench | 88.9% | estimated ± 2.6 pp, high confidence |
| HealthBench Hard | 12.3% | estimated ± 8.6 pp, low confidence |
| MedXpertQA (MM) | 76.3% | estimated ± 6.5 pp, low confidence |
| MMLU-Pro (Arcee) | 82.4% | estimated ± 4.0 pp, medium confidence |
| React Native Evals | 68.7% | estimated ± 3.0 pp, low confidence |
| ResearchClawBench | 12.5% | estimated ± 1.3 pp, medium confidence |
| SWE-Rebench | 32.8% | estimated ± 6.5 pp, low confidence |
| PostTrainBench | 17.2% | estimated ± 3.7 pp, medium confidence |
| DAYJOB | 0.0% | measured |
| DAYJOB Healthcare | 0.0% | measured |
| BioMysteryBench (human-solvable) | 81.7% | estimated ± 0.6 pp, low confidence |
| HealthBench (length-adjusted) | 54.1% | estimated ± 3.7 pp, medium confidence |
| HealthBench Professional | 56.7% | estimated ± 4.2 pp, low confidence |
| HLE-Verified | 15.3% | estimated ± 4.6 pp, low confidence |
| LABBench2 | 82.3% | estimated ± 1.8 pp, low confidence |
| ScarfBench | 45.9% | estimated ± 12.3 pp, low confidence |
| BlenderWorld (text-only) | 70.1% | estimated ± 2.8 pp, medium confidence |
| Cookie-Bench | 77.1% | estimated ± 1.3 pp, medium confidence |
| STEMBench | 2.5% | estimated ± 0.2 pp, medium confidence |
| SimpleQA | 53.5% | estimated ± 4.3 pp, low confidence |
| BabyVision w/ Python | 90.0% | estimated ± 4.0 pp, low confidence |
| OpenHarmony Bench v1.0 | 52.7% | estimated ± 3.2 pp, high confidence |
| SEC-Bench Pro | 80.8% | estimated ± 5.6 pp, medium confidence |
| MRCRv2 | 23.5% | estimated ± 11.3 pp, low confidence |
| MMLongBench-Doc | 59.9% | estimated ± 2.6 pp, low confidence |
| BioBasicsBench | 34.1% | estimated ± 3.8 pp, medium confidence |
| MMMU-Pro w/ Python | 81.5% | estimated ± 2.0 pp, high confidence |
| ExploitBench Internal Port | 0.4% | estimated ± 1.4 pp, medium confidence |