Google DeepMindGA
Gemini 3 Pro
Widest modality support with 2M native context.
Context
2M
Max output
—
Input /1M
—
Output /1M
—
Best for
- Multimodal understanding (81% MMMU-Pro)
- Video + audio + vision reasoning
- 2M-context tasks
Benchmarks
| mmmu pro | 81 |
| arc agi 2 | 45.1 |