Gemini 3 Pro
Released 2025-11 · hybrid · 1.0M tokens · 10 benchmarks
Editorial notes
Lanzado 18 noviembre 2025. Lidera LMArena en text/vision/coding/webdev. Variante Deep Think llega a 41.0% en HLE.
Spec sheet
- Company
- Google DeepMind
- Country
- US
- Type
- hybrid
- Release
- 2025-11
- Context
- 1.0M tokens
- Pricing (google)
- $2/$12/M
- Slug
- gemini-3-pro
Benchmarks (10)
Reasoning 5
- 91.9GPQA-DiamondGraduate-level Physics, Chemistry, Biology — PhD-level questions.
- 91.8MMLUMassive Multitask Language Understanding — 57 academic subjects, ~16K questions.
- 89.8MMLU-ProMMLU upgraded with harder questions and 10 answer options.
- 37.5Humanitys-Last-ExamThe hardest known benchmark — novel academic problems.
- 31.1ARC-AGI-2Updated ARC challenge — novel and tough abstract reasoning.
Coding 2
Knowledge 1
Cite this model
BibTeX · APA
BibTeX
@misc{frontier-gemini-3-pro,
title = {Gemini 3 Pro},
author = {{Google DeepMind}},
year = {2025},
note = {Frontier Benchmarks AI atlas. Accessed 2026-05-08},
url = {https://frontierbenchmarks.com/models/gemini-3-pro}
} APA
Google DeepMind (2025). Gemini 3 Pro [Large language model]. Frontier Benchmarks AI. Retrieved 2026-05-08, from https://frontierbenchmarks.com/models/gemini-3-pro
Citation reflects the atlas page, not the original model paper. For the paper, see the "Resources" section above.