Google · model
Gemini 4 Hydrogen benchmark scores
As of 2026-11-18, Gemini 4 Hydrogen (Google) has measured scores on 16 benchmarks and estimated scores on 70 more.
| Benchmark | Score | Source |
|---|---|---|
| AA-GPQA Diamond | 90.9% | estimated ± 11.5 pp, low confidence |
| AA-HLE | 53.7% | estimated ± 9.2 pp, medium confidence |
| AA-SciCode | 62.1% | estimated ± 2.5 pp, high confidence |
| Artificial Analysis Intelligence Index | 53.9% | measured |
| CritPt | 28.7% | estimated ± 5.3 pp, medium confidence |
| GDPval-AA | 59.2% | estimated ± 1.7 pp, medium confidence |
| GPQA Diamond | 92.9% | estimated ± 2.5 pp, high confidence |
| HLE | 52.9% | estimated ± 10.3 pp, low confidence |
| AA-Omniscience Accuracy | 50.5% | measured |
| Terminal-Bench 2.1 | 89.1% | estimated ± 5.4 pp, low confidence |
| BrowseComp | 90.9% | estimated ± 11.3 pp, low confidence |
| HLE w/ tools | 66.4% | estimated ± 6.8 pp, low confidence |
| AA Coding Index | 83.1% | estimated ± 1.2 pp, medium confidence |
| AA-MMMU-Pro | 86.6% | estimated ± 3.0 pp, medium confidence |
| BioMysteryBench (human-difficult) | 46.8% | estimated ± 7.8 pp, low confidence |
| AutomationBench | 50.6% | estimated ± 6.1 pp, low confidence |
| JobBench | 72.5% | estimated ± 11.3 pp, low confidence |
| SWE-bench Pro | 84.2% | estimated ± 4.2 pp, high confidence |
| MMLU-Pro | 85.5% | estimated ± 5.1 pp, low confidence |
| GPQA | 93.0% | estimated ± 6.9 pp, medium confidence |
| MMLU | 90.5% | estimated ± 0.9 pp, medium confidence |
| Vals GPQA Diamond | 92.7% | estimated ± 5.7 pp, medium confidence |
| Vals LiveCodeBench | 91.5% | estimated ± 3.0 pp, medium confidence |
| Vals MMLU-Pro | 88.4% | estimated ± 2.9 pp, high confidence |
| AA MMLU-Pro | 88.8% | estimated ± 0.8 pp, high confidence |
| AA-AnalystAgent | 53.9% | estimated ± 2.7 pp, medium confidence |
| ApprenticeBench | 15.4% | estimated ± 12.5 pp, low confidence |
| ARC-AGI-1 | 91.4% | estimated ± 7.9 pp, low confidence |
| ARC-AGI-2 | 83.7% | estimated ± 14.6 pp, low confidence |
| CursorBench 3.2 | 74.6% | estimated ± 2.0 pp, medium confidence |
| Terminal-Bench 3.0 | 44.2% | estimated ± 3.2 pp, medium confidence |
| FrontierCode 1.1 Main | 51.1% | estimated ± 3.8 pp, low confidence |
| FrontierSWE v2 | 53.1% | measured |
| MazeBench | 1.0% | measured |
| ParseBench-1 | 73.0% | estimated ± 5.7 pp, low confidence |
| Terminal-Bench 4.1 | 56.4% | measured |
| Terminal-Bench Hard | 68.1% | estimated ± 2.8 pp, low confidence |
| Vals SWE-bench | 100.0% | estimated ± 3.7 pp, medium confidence |
| Terminal-Bench 2.1 (Vals AI) | 80.3% | estimated ± 6.9 pp, low confidence |
| VulcanBench v3 | 87.6% | estimated ± 3.9 pp, medium confidence |
| CWE-Bench-AA | 62.9% | estimated ± 4.1 pp, medium confidence |
| DeepsecBench-AA | 31.8% | estimated ± 6.1 pp, low confidence |
| Harvey LAB-AA v1.1 | 8.9% | estimated ± 3.5 pp, medium confidence |
| HealthBench Professional (length-adjusted, Artificial Analysis) | 62.3% | estimated ± 4.3 pp, medium confidence |
| HealthBench Professional (raw, Artificial Analysis) | 69.9% | estimated ± 1.1 pp, medium confidence |
| Bug Hunt Bench | 32.7% | measured |
| Chartography (no tools) | 66.6% | measured |
| Chartography (tools) | 90.2% | estimated ± 2.3 pp, medium confidence |
| CursorBench 4.0 | 54.3% | measured |
| CWE-bench v1 | 57.3% | estimated ± 5.3 pp, medium confidence |
| DeepSWE | 74.0% | estimated ± 4.1 pp, medium confidence |
| DRACO | 1.7% | estimated ± 0.7 pp, low confidence |
| FrontierCode 2.0 | 38.9% | measured |
| HLE w/o tools | 59.0% | estimated ± 5.0 pp, high confidence |
| OSWorld 2.1 | 78.6% | estimated ± 6.1 pp, medium confidence |
| OSWorld 2.1 Offline | 77.6% | measured |
| PostTrainBench v1.1 | 48.6% | estimated ± 1.7 pp, medium confidence |
| SWE-bench Multimodal | 73.6% | estimated ± 8.3 pp, low confidence |
| Terminal-Bench 4.0 | 58.3% | estimated ± 3.7 pp, high confidence |
| Terminal-Bench-Science 0.1 | 51.4% | measured |
| Toolathlon-Verified | 78.0% | estimated ± 1.8 pp, low confidence |
| VulcanBench Frontier v4 | 92.2% | estimated ± 1.8 pp, high confidence |
| AA AutomationBench | 76.1% | measured |
| GDP.pdf | 27.6% | measured |
| AA Terminal-Bench 4.0 | 53.5% | measured |
| CWE-bench | 68.1% | measured |
| MKA | 53.4% | measured |
| Hermes Bench | 100.0% | estimated ± 1.8 pp, low confidence |
| Hermes Index | 88.9% | estimated ± 2.7 pp, low confidence |
| ParseBench-2 | 82.6% | estimated ± 6.4 pp, low confidence |
| ExploitGym | 13.9% | estimated ± 6.4 pp, medium confidence |
| MMLU-Redux | 97.3% | estimated ± 0.8 pp, low confidence |
| NL2Repo | 64.6% | estimated ± 6.5 pp, low confidence |
| SuperGPQA | 99.9% | estimated ± 9.3 pp, low confidence |
| PostTrainBench | 37.5% | estimated ± 2.4 pp, high confidence |
| Finance Agent v2 | 56.2% | estimated ± 3.8 pp, medium confidence |
| FrontierCode 1.1 Extended | 62.1% | estimated ± 3.0 pp, medium confidence |
| HealthBench (length-adjusted) | 60.4% | estimated ± 3.7 pp, high confidence |
| HealthBench Professional | 62.3% | estimated ± 4.2 pp, medium confidence |
| HLE-Verified | 46.9% | estimated ± 7.1 pp, low confidence |
| ScarfBench | 69.3% | estimated ± 4.5 pp, low confidence |
| STEMBench | 2.6% | estimated ± 0.2 pp, medium confidence |
| Agents' Last Exam | 50.1% | estimated ± 10.7 pp, low confidence |
| OpenHarmony Bench v1.2 | 63.8% | estimated ± 11.0 pp, low confidence |
| SWE-Marathon | 51.2% | estimated ± 2.1 pp, medium confidence |
| BioBasicsBench | 34.1% | estimated ± 3.8 pp, medium confidence |