Google · model
Gemini 3.8 Flash benchmark scores
As of 2026-10-07, Gemini 3.8 Flash (Google) has measured scores on 38 benchmarks and estimated scores on 73 more.
| Benchmark | Score | Source |
|---|---|---|
| AA-GPQA Diamond | 95.3% | measured |
| AA-HLE | 47.8% | measured |
| AA-SciCode | 56.6% | measured |
| Artificial Analysis Intelligence Index | 40.9% | measured |
| CritPt | 18.3% | measured |
| GDPval-AA | 46.8% | measured |
| GPQA Diamond | 95.2% | estimated ± 1.4 pp, high confidence |
| HLE | 60.7% | estimated ± 5.5 pp, low confidence |
| AA-LCR | 81.3% | measured |
| LiveCodeBench v6 | 93.5% | estimated ± 3.1 pp, medium confidence |
| AA-Omniscience Accuracy | 54.6% | measured |
| SciCode | 55.8% | estimated ± 2.9 pp, low confidence |
| Terminal-Bench 2.1 | 89.4% | measured |
| BrowseComp | 89.4% | estimated ± 2.2 pp, high confidence |
| HLE w/ tools | 55.4% | estimated ± 4.6 pp, low confidence |
| LongBench v2 | 61.5% | estimated ± 5.9 pp, low confidence |
| τ²-bench results | 85.4% | estimated ± 4.8 pp, medium confidence |
| AA Coding Index | 76.3% | measured |
| AA-MMMU-Pro | 85.6% | measured |
| APEX-Agents | 39.0% | estimated ± 10.0 pp, low confidence |
| APEX-Agents-AA | 38.7% | estimated ± 2.9 pp, low confidence |
| BioMysteryBench (human-difficult) | 56.5% | measured |
| DeepSearchQA | 91.9% | estimated ± 4.5 pp, high confidence |
| SWE-bench Verified | 95.8% | estimated ± 4.4 pp, high confidence |
| AutomationBench | 43.8% | estimated ± 7.6 pp, low confidence |
| BFCL v4 | 77.2% | estimated ± 2.6 pp, medium confidence |
| CyberGym | 85.6% | estimated ± 5.0 pp, medium confidence |
| JobBench | 64.6% | estimated ± 4.9 pp, medium confidence |
| SWE-bench Pro | 67.4% | estimated ± 4.3 pp, high confidence |
| WideResearch | 97.0% | estimated ± 9.1 pp, low confidence |
| AA Agentic Index | 41.1% | measured |
| MMLU-Pro | 85.7% | estimated ± 3.5 pp, medium confidence |
| GPQA | 94.8% | estimated ± 0.8 pp, high confidence |
| Gert Labs | 74.4% | estimated ± 3.0 pp, low confidence |
| AA-AnalystAgent | 54.4% | estimated ± 5.5 pp, medium confidence |
| AA EnterpriseOps-Gym | 46.1% | estimated ± 4.4 pp, high confidence |
| AA Harvey LAB | 92.4% | estimated ± 1.2 pp, medium confidence |
| ApprenticeBench | 24.0% | measured |
| ARC-AGI-1 | 98.5% | measured |
| ARC-AGI-2 | 89.2% | measured |
| CursorBench 3.1 | 60.3% | estimated ± 3.4 pp, medium confidence |
| CursorBench 3.2 | 69.2% | measured |
| Terminal-Bench 3.0 | 28.3% | estimated ± 6.9 pp, medium confidence |
| FrontierCode 1.1 Main | 52.8% | estimated ± 1.6 pp, medium confidence |
| FrontierSWE v2 | 19.6% | measured |
| OfficeQA Pro | 67.7% | estimated ± 7.2 pp, medium confidence |
| OSWorld-Verified | 81.3% | estimated ± 1.9 pp, medium confidence |
| Vals GPQA Diamond | 94.4% | measured |
| Vals LiveCodeBench | 89.5% | measured |
| Vals MMLU-Pro | 90.2% | measured |
| Vals SWE-bench | 80.0% | measured |
| Terminal-Bench 2.1 (Vals AI) | 81.3% | measured |
| VulcanBench v3 | 87.0% | estimated ± 0.7 pp, medium confidence |
| AA AutomationBench | 59.9% | measured |
| GDP.pdf | 21.0% | measured |
| AA ITBench | 52.5% | measured |
| AA Tau3 Banking | 44.9% | measured |
| AA Terminal-Bench 2.1 | 87.6% | measured |
| AA Terminal-Bench 4.0 | 19.7% | measured |
| CursorBench 4.0 | 39.6% | measured |
| CWE-bench v1 | 56.2% | estimated ± 6.0 pp, medium confidence |
| DeepSWE | 73.8% | measured |
| HLE w/o tools | 52.5% | estimated ± 5.0 pp, medium confidence |
| OSWorld 2.0 | 59.0% | measured |
| PostTrainBench v1.1 | 23.6% | estimated ± 2.7 pp, high confidence |
| ProgramBench | 88.0% | estimated ± 6.0 pp, low confidence |
| Terminal-Bench 4.0 | 19.1% | measured |
| Terminal-Bench-Science 0.1 | 0.8% | estimated ± 4.1 pp, low confidence |
| Toolathlon-Verified | 79.3% | estimated ± 1.0 pp, medium confidence |
| Vibe Code Bench | 72.9% | estimated ± 10.3 pp, low confidence |
| CharXiv | 91.4% | estimated ± 2.5 pp, high confidence |
| CharXiv w/o tools | 86.2% | measured |
| ExploitGym | 16.6% | estimated ± 5.0 pp, low confidence |
| C-Eval | 95.8% | estimated ± 1.0 pp, low confidence |
| Claw-Eval | 80.2% | estimated ± 13.6 pp, low confidence |
| MathVision | 93.5% | estimated ± 2.1 pp, medium confidence |
| MCP Atlas | 85.5% | estimated ± 2.7 pp, low confidence |
| MMLU-Redux | 97.6% | estimated ± 1.1 pp, low confidence |
| MMMU-Pro | 82.2% | estimated ± 1.4 pp, medium confidence |
| NL2Repo | 45.9% | estimated ± 4.7 pp, high confidence |
| QwenClawBench | 100.0% | estimated ± 6.0 pp, low confidence |
| ScreenSpot Pro | 85.4% | estimated ± 10.7 pp, low confidence |
| SuperGPQA | 87.3% | estimated ± 7.5 pp, low confidence |
| Terminal-Bench 2.0 | 80.6% | estimated ± 5.9 pp, low confidence |
| Toolathlon | 56.7% | estimated ± 1.7 pp, medium confidence |
| ERQA | 68.9% | estimated ± 3.9 pp, medium confidence |
| HealthBench Hard | 33.7% | estimated ± 7.7 pp, low confidence |
| MedXpertQA (MM) | 77.8% | estimated ± 6.5 pp, low confidence |
| MMLU-Pro (Arcee) | 100.0% | estimated ± 4.0 pp, low confidence |
| React Native Evals | 81.3% | estimated ± 2.4 pp, medium confidence |
| ResearchClawBench | 20.2% | estimated ± 2.0 pp, high confidence |
| SWE-Rebench | 58.1% | estimated ± 8.9 pp, low confidence |
| ARC-AGI-3 | 10.4% | measured |
| BioMysteryBench (human-solvable) | 88.8% | measured |
| DRACO | 80.7% | estimated ± 3.0 pp, medium confidence |
| FrontierCode 1.1 Extended | 62.3% | estimated ± 1.6 pp, medium confidence |
| HealthBench (raw) | 56.0% | estimated ± 6.4 pp, low confidence |
| HealthBench (length-adjusted) | 53.9% | estimated ± 0.9 pp, medium confidence |
| HealthBench Professional | 60.0% | estimated ± 1.3 pp, medium confidence |
| HealthBench Professional (raw) | 65.9% | estimated ± 5.0 pp, medium confidence |
| HLE-Verified | 54.9% | measured |
| LABBench2 | 86.2% | measured |
| MMMU | 92.2% | estimated ± 0.9 pp, medium confidence |
| LiveCodeBench | 100.0% | estimated ± 8.9 pp, low confidence |
| Agents' Last Exam | 42.9% | estimated ± 6.3 pp, medium confidence |
| OpenHarmony Bench | 58.8% | estimated ± 2.4 pp, high confidence |
| MMLU | 95.5% | estimated ± 1.1 pp, low confidence |
| MMMU-Pro w/ Python | 85.0% | estimated ± 1.6 pp, medium confidence |
| MathVision w/ Python | 97.0% | estimated ± 1.3 pp, low confidence |
| RealWorldQA | 90.4% | estimated ± 1.2 pp, low confidence |
| OmniDocBench 1.5 | 92.0% | estimated ± 6.4 pp, low confidence |