GPT-5 is OpenAI's unified frontier model released in mid-2025, bringing together advanced reasoning, coding, and multimodal capabilities into a single system. It introduced test-time compute scaling with configurable thinking depth, significantly reducing hallucinations and sycophancy compared to previous models. GPT-5 excels at complex multi-step tasks requiring step-by-step reasoning, instruction following, and accuracy in high-stakes scenarios, with notable improvements in coding, writing, and factual reliability.
API|VisionReasoningFile|Proprietary Model
Knowledge Cutoff
2024-09-30
Input → Output Format
Context Memory
400KIN128KOUT
AI Performance Evaluation
Arena Overall Score
1434
±5As of 2026-04-23
Overall Rank
No.54
31,986 Votes
Arena by Ability
Hard Prompts
1446±6No.65
Expert Knowledge
1459±16No.47
Instruction Following
1409±7No.72
Conversation Memory
1420±9No.73
Creative
1375±10No.96
Coding
1466±8No.67
Math
1434±14No.46
Arena by Occupation
Creative Writing
1397±8No.76
Social Sciences
1443±9No.66
Media
1397±8No.60
Business
1414±9No.76
Healthcare
1456±15No.56
Legal
1455±14No.45
Software
1452±7No.78
Mathematics
1441±14No.49
Source:Arena Intelligence
Overall
AA Intelligence Index
22%↓17%
LiveBench
71%↑11%
ForecastBench
61%↑2%
Reasoning & Math
AA Math Index
48%↓25%
GPQA Diamond
69%↓12%
HLE
5.8%↓11%
MMLU-Pro
82%↑0%
AIME 2025
48%↓25%
LB Reasoning
82%↑22%
LB Math
86%↑13%
LB Data
57%↑7%
Coding
AA Coding Index
21%↓13%
LiveCodeBench
54%↓11%
LB Coding
72%↓2%
LB Agentic
52%↑8%
TAU2
0.0%↓73%
TerminalBench
13%↓18%
SciCode
38%↓3%
Language & Instructions
IFBench
45%↓12%
AA-LCR
64%↑2%
Hallucination (HHEM)
15%↑5%
Factual (HHEM)
85%↓5%
LB Language
81%↑9%
LB IF
64%↑18%
Output Speed
Standard Mode
77tok/s↓5
First Output 1.03s
Reasoning Mode
85tok/s↓3
First Output 41.72s
