Google
Gemini 4 Hydrogen
Released · blog.google · Launch vibes
Context window
| Tokens | Context window | Input limit | Output limit |
|---|
| Defaultwindow is a unified input and output limit; default output is for a single foreground segment | 1,048,576 | 1M | 65,536 |
| With long decode | – | – | 1M |
Context history
Benchmark results
| Benchmark | Score |
|---|
| HLE-DiamondNo toolsRun by Artificial Analysis · highest thinking · Setting from Artificial Analysis’s methodology | 58.8%Corroborated The same number is printed in 2 other lab documents: 1 2 |
| Terminal-Bench Science 0.1mini-swe-agent vision harness · highest thinking · average of 5 attempts per task | 51.4%Corroborated The same number is printed in 3 other lab documents: 1 2 3 |
| AutomationBench 1.0.6AutomationBench-AARun by Artificial Analysis · highest thinking | 76.1%Corroborated The same number is printed in 3 other lab documents: 1 2 3 |
| OSWorld 2.1Offline set (82 tasks), partial scoreGemini CUA harness · highest thinking · max of 3 runs, one attempt per run | 77.6%Corroborated The same number is printed in 3 other lab documents: 1 2 3 |
| ChartographyNo toolshighest thinking | 66.6%Corroborated The same number is printed in 3 other lab documents: 1 2 3 |
| Agents' Last Exam v1.1Score, all tasks (122)Antigravity-CLI (agy) version 1.4.8 harness · highest thinking · 5-hour window · October 17, 2026 task files · safety filters on | 71.6%Corroborated The same number is printed in 3 other lab documents: 1 2 3 |
Benchmark sources
Reported by Google
All models