Google DeepMind · closed · 2025-11
Gemini 3 Pro
#5 index 86.5
the takeThe model that opened the generation, and a fair baseline. It is beaten by 3.1 Pro on every axis that counts, so buy the newer one unless price forces your hand.
SWE-bench Verified
76.2
GPQA Diamond
91.9
MMLU-Pro
89.8
AIME (math)
95.0
- Context
- 1.048576M
- Input
- $2/M
- Output
- $12/M
- Speed
- —
- Modality
- text, image, audio, video
- Strong all-round reasoning and math
- Complete, documented benchmark card
- Broadly integrated
- Superseded by 3.1 Pro
- Weak abstract reasoning (ARC-AGI-2 31%)
- Long-context surcharge