ModelBeat
Models/GPT-5.4
All models
O

OpenAI: GPT-5.4

openai/gpt-5.4

GPT-5.4 is OpenAI’s latest frontier model, unifying the Codex and GPT lines into a single system.

Context
1.1M
Input
$2.50 / 1M
Output
$15 / 1M
Released
Mar 5, 2026
Modalities
textimagefiletext

Benchmarks

Scores on standardized evaluations. Higher is better — and rank shows where GPT-5.4 lands among all models tracked on Model Beat.

83.0
Intelligence Index
Epoch AI
83rd percentile of tracked models
80.0
Coding Index
Epoch AI
80th percentile of tracked models
84.0
Agentic Index
Epoch AI
84th percentile of tracked models

Reasoning

6 evals
ARC-AGI93.7%

Abstract visual reasoning

ARC-AGI-274.0%

Harder abstract reasoning

GPQA Diamond93.3%

Graduate-level scientific reasoning

Humanity's Last Exam36.2%

Frontier of human expert knowledge

SimpleQA Verified44.8%

Factual accuracy & hallucination

WeirdML77.7%

Novel ML problem-solving

Coding

3 evals
GSO (code optimization)31.4%

Code performance optimization

SWE-bench Verified76.9%

Real GitHub issue resolution

WebDev Arena1457

Human-rated web development

Math

3 evals
AIME 2024/202595.3%

Olympiad-qualifier math

FrontierMath47.6%

Research-level math problems

FrontierMath Tier 427.1%

Hardest research math

Agentic & Tools

3 evals
APEX35.9%

Multi-step agentic tasks

METR task horizon5.7 h

Autonomous task length

Terminal-Bench81.8%

Command-line agentic tasks

In the news

6

Model & benchmark data from Epoch AI (CC BY); pricing, specs & descriptions from OpenRouter.