Gemini 3.1 Pro is Google's most advanced reasoning model, significantly outperforming Gemini 3 Pro across software engineering, agentic reliability, and token efficiency. It supports a 1M-token context window with multimodal inputs including text, images, video, audio, code, and PDFs, and introduces a new medium thinking level for better cost-speed-performance balance. The model excels at agentic coding, structured planning, financial modeling, spreadsheet automation, and high-context enterprise tasks requiring long-horizon stability and autonomous tool orchestration.
Google AI PlusGoogle AI ProGoogle AI UltraAPI|VisionReasoningWeb SearchFile|Proprietary Model
Knowledge Cutoff
2025-01-31
Input → Output Format
Context Memory
1.0MIN66KOUT
AI Performance Evaluation
Arena Overall Score
1493
±5As of 2026-04-23
Overall Rank
No.5
25,353 Votes
Arena by Ability
Hard Prompts
1515±6No.5
Expert Knowledge
1520±14No.6
Instruction Following
1489±7No.5
Conversation Memory
1505±10No.6
Creative
1488±10🥉 No.3
Coding
1531±8No.7
Math
1509±15🥉 No.3
Arena by Occupation
Creative Writing
1487±8🥉 No.3
Social Sciences
1513±10🥉 No.3
Media
1476±9🥉 No.3
Business
1483±9No.7
Healthcare
1510±14No.9
Legal
1505±14No.4
Software
1520±7No.6
Mathematics
1497±16No.5
Source:Arena Intelligence
Overall
AA Intelligence Index
57%↑19%
LiveBench
81%↑20%
ForecastBench
60%↑1%
Reasoning & Math
GPQA Diamond
94%↑13%
HLE
45%↑28%
LB Reasoning
84%↑24%
LB Math
91%↑17%
LB Data
79%↑29%
Coding
AA Coding Index
56%↑21%
LB Coding
76%↑3%
LB Agentic
65%↑22%
TAU2
96%↑22%
TerminalBench
54%↑23%
SciCode
59%↑18%
Language & Instructions
IFBench
77%↑20%
AA-LCR
73%↑11%
Hallucination (HHEM)
10%↑0%
Factual (HHEM)
90%↑0%
LB Language
85%↑14%
LB IF
79%↑33%
Output Speed
Standard Mode
135tok/s↑53
First Output 22.87s
Multilingual Capabilities
MGSM 🇰🇷
94%
MGSM 🇯🇵
94%
KMMLU 🇰🇷
82%
JMMLU 🇯🇵
82%
