GPT-5.4 Mini brings the core capabilities of GPT-5.4 to a faster, more efficient form factor optimized for high-throughput workloads. It runs over 2× faster than GPT-5 Mini while approaching GPT-5.4's performance on coding and reasoning benchmarks, and supports text and image inputs with full tool use, web search, and function calling. With a 400K-token context window, it delivers reliable instruction following and multi-step reasoning at significantly reduced cost, making it well-suited for chat applications, coding assistants, and agent workflows operating at scale.
API|VisionReasoningWeb SearchFile|Proprietary Model
Knowledge Cutoff
2025-08-31
Input → Output Format
Context Memory
400KIN128KOUT
AI Performance Evaluation
Arena Overall Score
1456
±6As of 2026-05-01
Overall Rank
No.33
13,541 Votes
Arena by Ability
Hard Prompts
1479±7No.32
Expert Knowledge
1489±17No.24
Instruction Following
1441±9No.39
Conversation Memory
1474±12No.27
Creative
1411±13No.50
Coding
1508±11No.28
Math
1437±20No.50
Arena by Occupation
Creative Writing
1432±11No.40
Social Sciences
1468±12No.38
Media
1422±12No.41
Business
1470±12No.19
Healthcare
1455±20No.63
Legal
1459±19No.41
Software
1494±9No.30
Mathematics
1460±21No.35
Source:Arena Intelligence
Overall
AA Intelligence Index
38%↓1%
LiveBench
34%↓27%
ForecastBench
56%↓3%
Reasoning & Math
GPQA Diamond
82%↑0%
HLE
17%↑0%
LB Reasoning
22%↓47%
LB Math
37%↓37%
LB Data
47%↓6%
Coding
AA Coding Index
38%↑1%
LB Coding
75%↑2%
LB Agentic
17%↓28%
TAU2
37%↓44%
TerminalBench
34%↑0%
SciCode
44%↑2%
Language & Instructions
IFBench
65%↑2%
AA-LCR
61%↓1%
Hallucination (HHEM)
5.5%↓5%
Factual (HHEM)
95%↑5%
LB Language
42%↓30%
LB IF
19%↓32%
Output Speed
Standard Mode
172tok/s↑95
First Output 0.48s
Reasoning Mode
170tok/s↑83
First Output 5.26s
