GPT-5 Mini is a compact variant of GPT-5, designed for lighter-weight reasoning tasks with reduced latency and cost. As the successor to o4-mini, it retains GPT-5's instruction-following and safety capabilities while being optimized for everyday workloads that don't require the full model's depth. It offers a practical balance of reasoning ability and efficiency for applications where responsiveness and cost management are key considerations.
OpenAI FreeOpenAI GoOpenAI PlusOpenAI ProAPI|VisionReasoningWeb SearchFile|Proprietary Model
Knowledge Cutoff
2024-05-31
Input → Output Format
Context Memory
400KIN128KOUT
AI Performance Evaluation
Arena Overall Score
1390
±5As of 2026-04-23
Overall Rank
No.113
27,067 Votes
Arena by Ability
Hard Prompts
1402±6No.120
Expert Knowledge
1403±18No.110
Instruction Following
1374±8No.120
Conversation Memory
1372±9No.132
Creative
1325±10No.145
Coding
1429±9No.123
Math
1406±15No.91
Arena by Occupation
Creative Writing
1353±8No.134
Social Sciences
1396±10No.125
Media
1349±9No.119
Business
1380±9No.123
Healthcare
1401±17No.121
Legal
1394±15No.121
Software
1415±7No.126
Mathematics
1409±17No.100
Source:Arena Intelligence
Overall
AA Intelligence Index
21%↓18%
LiveBench
61%↑1%
ForecastBench
60%↑1%
Reasoning & Math
AA Math Index
47%↓27%
GPQA Diamond
69%↓12%
HLE
5.0%↓12%
MMLU-Pro
78%↓4%
AIME 2025
47%↓27%
LB Reasoning
59%↓1%
LB Math
74%↑1%
LB Data
50%↑0%
Coding
AA Coding Index
22%↓12%
LiveCodeBench
55%↓11%
LB Coding
76%↑2%
LB Agentic
35%↓8%
TAU2
32%↓41%
TerminalBench
14%↓17%
SciCode
37%↓4%
Language & Instructions
IFBench
46%↓11%
AA-LCR
36%↓26%
Hallucination (HHEM)
13%↑3%
Factual (HHEM)
87%↓3%
LB Language
69%↓3%
LB IF
64%↑18%
Output Speed
Standard Mode
79tok/s↓3
First Output 0.95s
Reasoning Mode
88tok/s↑0
First Output 48.03s
