Google · model
Gemini 3.5 Flash benchmark scores
As of 2026-10-07, Gemini 3.5 Flash (Google) has measured scores on 41 benchmarks and estimated scores on 86 more.
| Benchmark | Score | Source |
|---|---|---|
| AA-GPQA Diamond | 92.2% | measured |
| AA-HLE | 42.7% | measured |
| AA-SciCode | 53.9% | measured |
| AIME26 | 96.4% | estimated ± 0.6 pp, low confidence |
| Artificial Analysis Intelligence Index | 50.2% | measured |
| CritPt | 13.1% | measured |
| GDPval-AA | 42.2% | measured |
| GPQA Diamond | 92.7% | measured |
| HLE | 40.2% | measured |
| IFBench | 76.3% | measured |
| AA-LCR | 69.3% | measured |
| LiveCodeBench v6 | 87.4% | estimated ± 2.1 pp, high confidence |
| AA-Omniscience Accuracy | 51.9% | measured |
| SciCode | 53.1% | measured |
| Terminal-Bench 2.1 | 76.2% | measured |
| BrowseComp | 83.8% | estimated ± 3.1 pp, high confidence |
| HLE w/ tools | 53.2% | estimated ± 6.1 pp, medium confidence |
| IFEval | 93.7% | estimated ± 1.3 pp, high confidence |
| LongBench v2 | 61.5% | estimated ± 5.9 pp, medium confidence |
| VITA-Bench | 49.6% | estimated ± 9.9 pp, low confidence |
| τ²-bench results | 95.3% | measured |
| AA Coding Index | 70.1% | measured |
| AA-MMMU-Pro | 84.3% | measured |
| APEX-Agents | 36.8% | estimated ± 0.6 pp, medium confidence |
| APEX-Agents-AA | 47.1% | measured |
| BioMysteryBench (human-difficult) | 46.4% | estimated ± 5.9 pp, low confidence |
| DeepSearchQA | 88.1% | estimated ± 4.5 pp, high confidence |
| SWE-bench Verified | 78.6% | estimated ± 2.5 pp, high confidence |
| AutomationBench | 33.2% | estimated ± 7.0 pp, low confidence |
| BFCL v4 | 76.8% | estimated ± 2.6 pp, medium confidence |
| CyberGym | 66.5% | estimated ± 5.3 pp, medium confidence |
| JobBench | 54.3% | estimated ± 6.8 pp, medium confidence |
| SWE-bench Pro | 55.1% | measured |
| τ³-bench results | 69.6% | estimated ± 1.8 pp, low confidence |
| WideResearch | 84.7% | estimated ± 5.6 pp, low confidence |
| AA Agentic Index | 27.3% | measured |
| MMLU-Pro | 86.8% | estimated ± 3.0 pp, high confidence |
| FrontierMath v2 (Tier 4) | 14.6% | measured |
| FrontierMath v2 (Tiers 1-3) | 39.0% | measured |
| GPQA | 92.2% | measured |
| AA-IFBench | 76.3% | measured |
| Gert Labs | 61.9% | measured |
| AA-AnalystAgent | 45.2% | estimated ± 3.6 pp, medium confidence |
| AA EnterpriseOps-Gym | 50.1% | measured |
| AA Harvey LAB | 85.6% | estimated ± 1.2 pp, medium confidence |
| ApprenticeBench | 3.3% | estimated ± 3.1 pp, medium confidence |
| ARC-AGI-1 | 92.5% | measured |
| ARC-AGI-2 | 72.1% | measured |
| CursorBench 3.1 | 49.8% | measured |
| CursorBench 3.2 | 48.8% | measured |
| Terminal-Bench 3.0 | 18.2% | estimated ± 6.9 pp, medium confidence |
| FrontierCode 1.1 Main | 32.2% | estimated ± 1.6 pp, low confidence |
| FrontierSWE v2 | 6.4% | estimated ± 9.0 pp, medium confidence |
| OfficeQA Pro | 67.7% | estimated ± 7.2 pp, medium confidence |
| OSWorld-Verified | 78.4% | measured |
| Vals GPQA Diamond | 92.7% | measured |
| Vals LiveCodeBench | 87.6% | measured |
| Vals MMLU-Pro | 89.5% | measured |
| Vals SWE-bench | 78.8% | measured |
| Terminal-Bench 2.1 (Vals AI) | 74.2% | measured |
| VulcanBench v3 | 71.2% | estimated ± 5.1 pp, low confidence |
| AA AutomationBench | 39.8% | estimated ± 7.0 pp, medium confidence |
| GDP.pdf | 17.3% | estimated ± 3.7 pp, low confidence |
| AA ITBench | 51.3% | estimated ± 3.6 pp, medium confidence |
| AA Tau3 Banking | 26.0% | estimated ± 5.8 pp, medium confidence |
| AA Terminal-Bench 2.1 | 83.5% | estimated ± 3.5 pp, high confidence |
| AA Terminal-Bench 4.0 | 14.1% | estimated ± 8.9 pp, medium confidence |
| CursorBench 4.0 | 32.7% | estimated ± 1.6 pp, medium confidence |
| CWE-bench v1 | 50.3% | estimated ± 6.9 pp, medium confidence |
| DeepSWE | 55.8% | estimated ± 6.4 pp, medium confidence |
| HLE w/o tools | 39.0% | estimated ± 2.9 pp, high confidence |
| OSWorld 2.0 | 13.0% | estimated ± 3.6 pp, high confidence |
| PostTrainBench v1.1 | 23.4% | estimated ± 2.7 pp, high confidence |
| ProgramBench | 50.6% | estimated ± 6.0 pp, low confidence |
| Toolathlon-Verified | 57.0% | estimated ± 2.3 pp, medium confidence |
| Vibe Code Bench | 48.7% | measured |
| CharXiv | 84.2% | measured |
| CharXiv w/o tools | 88.3% | estimated ± 1.8 pp, low confidence |
| ExploitGym | 1.1% | estimated ± 12.2 pp, low confidence |
| C-Eval | 93.2% | estimated ± 0.8 pp, low confidence |
| Claw-Eval | 63.7% | estimated ± 3.2 pp, high confidence |
| DeepPlanning | 51.7% | estimated ± 13.1 pp, low confidence |
| HMMT Nov 2025 | 93.2% | estimated ± 2.6 pp, low confidence |
| MathVision | 88.7% | estimated ± 2.6 pp, high confidence |
| MCP Atlas | 83.6% | measured |
| MCP-Tasks | 77.1% | estimated ± 0.6 pp, low confidence |
| MMAnswerBench | 85.2% | estimated ± 1.3 pp, low confidence |
| MMLU-Redux | 97.5% | estimated ± 1.1 pp, low confidence |
| MMMU-Pro | 83.6% | measured |
| NL2Repo | 45.1% | estimated ± 4.7 pp, high confidence |
| QwenClawBench | 68.9% | estimated ± 2.6 pp, medium confidence |
| ScreenSpot Pro | 78.7% | estimated ± 7.1 pp, medium confidence |
| SuperGPQA | 86.4% | estimated ± 7.5 pp, low confidence |
| Terminal-Bench 2.0 | 76.2% | estimated ± 5.9 pp, medium confidence |
| Toolathlon | 56.5% | measured |
| VideoMMMU | 89.2% | estimated ± 1.0 pp, medium confidence |
| ERQA | 68.2% | estimated ± 3.7 pp, high confidence |
| HealthBench Hard | 30.2% | estimated ± 5.1 pp, low confidence |
| LiveCodeBench Pro | 79.8% | estimated ± 6.0 pp, low confidence |
| MedXpertQA (MM) | 81.7% | estimated ± 2.4 pp, high confidence |
| MedXpertQA (Text) | 62.7% | estimated ± 4.5 pp, medium confidence |
| MMLU-Pro (Arcee) | 87.8% | estimated ± 3.4 pp, low confidence |
| React Native Evals | 79.3% | estimated ± 2.4 pp, medium confidence |
| ResearchClawBench | 18.0% | measured |
| SWE-Rebench | 62.0% | estimated ± 6.5 pp, low confidence |
| ARC-AGI-3 | 0.1% | estimated ± 2.1 pp, medium confidence |
| FrontierMath (legacy) | 39.6% | estimated ± 0.7 pp, low confidence |
| BioMysteryBench (human-solvable) | 86.1% | estimated ± 0.6 pp, low confidence |
| DRACO | 78.9% | estimated ± 3.0 pp, medium confidence |
| FrontierCode 1.1 Extended | 55.1% | estimated ± 1.6 pp, low confidence |
| HealthBench (raw) | 51.5% | estimated ± 6.4 pp, low confidence |
| HealthBench (length-adjusted) | 56.1% | estimated ± 2.5 pp, low confidence |
| HealthBench Professional | 59.9% | estimated ± 1.3 pp, medium confidence |
| HealthBench Professional (raw) | 61.7% | estimated ± 5.0 pp, medium confidence |
| HLE-Verified | 35.3% | estimated ± 0.7 pp, low confidence |
| LABBench2 | 75.9% | estimated ± 1.3 pp, low confidence |
| MMMU | 91.0% | estimated ± 0.9 pp, medium confidence |
| LiveCodeBench | 100.0% | estimated ± 8.9 pp, low confidence |
| OpenHarmony Bench | 58.8% | estimated ± 2.4 pp, high confidence |
| SimpleVQA | 68.8% | estimated ± 7.8 pp, low confidence |
| MMMLU | 90.7% | estimated ± 1.1 pp, low confidence |
| MMLU | 95.2% | estimated ± 1.1 pp, low confidence |
| MMMU-Pro w/ Python | 85.0% | estimated ± 0.6 pp, medium confidence |
| MathVision w/ Python | 76.8% | estimated ± 0.9 pp, low confidence |
| RealWorldQA | 86.1% | estimated ± 1.1 pp, medium confidence |
| Video-MME (with subtitle) | 88.3% | estimated ± 0.9 pp, medium confidence |
| OmniDocBench 1.5 | 91.7% | estimated ± 5.9 pp, low confidence |