Gemini 2.5 Pro is Google's state-of-the-art reasoning model, designed for advanced coding, mathematics, and scientific tasks that demand deep analytical thinking. It employs built-in "thinking" capabilities that enable step-by-step reasoning through complex problems with enhanced accuracy, and achieved first place on the LMArena leaderboard upon release, reflecting superior human-preference alignment. With a 1M-token context window and multimodal input support, it excels at complex problem-solving, long-document analysis, and research-grade workflows requiring the highest level of reasoning depth.
API|VisionReasoningWeb SearchFile|Proprietary Model
Knowledge Cutoff
2025-01-31
Input → Output Format
Context Memory
1.0MIN66KOUT
AI Performance Evaluation
Arena Overall Score
1448
±3As of 2026-05-01
Overall Rank
No.45
113,545 Votes
Arena by Ability
Hard Prompts
1460±3No.54
Expert Knowledge
1464±8No.50
Instruction Following
1441±4No.38
Conversation Memory
1449±5No.50
Creative
1447±5No.18
Coding
1465±5No.78
Math
1443±7No.42
Arena by Occupation
Creative Writing
1448±5No.26
Social Sciences
1472±5No.33
Media
1433±5No.30
Business
1437±5No.56
Healthcare
1468±8No.47
Legal
1467±7No.32
Software
1461±4No.67
Mathematics
1450±8No.41
Source:Arena Intelligence
Overall
AA Intelligence Index
35%↓5%
LiveBench
57%↓3%
ForecastBench
60%↑1%
Reasoning & Math
AA Math Index
88%↑13%
GPQA Diamond
84%↑2%
HLE
21%↑4%
MMLU-Pro
86%↑5%
AIME 2025
88%↑13%
MATH-500
97%↑4%
LB Reasoning
71%↑2%
LB Math
68%↓6%
LB Data
52%↓2%
Coding
AA Coding Index
32%↓5%
LiveCodeBench
80%↑15%
LB Coding
76%↑3%
LB Agentic
33%↓12%
TAU2
54%↓26%
TerminalBench
27%↓8%
SciCode
43%↑1%
Language & Instructions
IFBench
49%↓14%
AA-LCR
66%↑4%
Hallucination (HHEM)
7.0%↓3%
Factual (HHEM)
93%↑3%
LB Language
76%↑3%
LB IF
33%↓18%
Output Speed
Standard Mode
122tok/s↑45
First Output 17.99s
