ModelBeat
Models/GPT-5
All models
O

OpenAI: GPT-5

CompareAnnouncement
openai/gpt-5

GPT-5 is OpenAI’s most advanced model, offering major improvements in reasoning, code quality, and user experience.

Context
400K
Input
$1.25 / 1M
Output
$10 / 1M
Released
Aug 7, 2025
Modalities
textimagefile→text

Benchmarks

Scores on standardized evaluations. Higher is better — and rank shows where GPT-5 lands among all models tracked on Model Beat.

51.0
Intelligence Index
Epoch AI
51st percentile of tracked models
39.0
Coding Index
Epoch AI
39th percentile of tracked models
42.0
Agentic Index
Epoch AI
42nd percentile of tracked models

Reasoning

7 evals
ARC-AGI65.7%

Abstract visual reasoning

ARC-AGI-29.9%

Harder abstract reasoning

GPQA Diamond86.2%

Graduate-level scientific reasoning

Humanity's Last Exam25.3%

Frontier of human expert knowledge

SimpleBench56.7%

Common-sense trick questions

SimpleQA Verified50.6%

Factual accuracy & hallucination

WeirdML60.7%

Novel ML problem-solving

Coding

3 evals
GSO (code optimization)6.9%

Code performance optimization

SWE-bench Verified73.6%

Real GitHub issue resolution

WebDev Arena1395

Human-rated web development

Math

4 evals
AIME 2024/202591.4%

Olympiad-qualifier math

FrontierMath32.4%

Research-level math problems

FrontierMath Tier 412.5%

Hardest research math

MATH Level 598.1%

Hardest competition math

Agentic & Tools

4 evals
APEX18.3%

Multi-step agentic tasks

GDPval (win/tie rate)38.0%

Economically valuable work

METR task horizon3.4 h

Autonomous task length

Terminal-Bench49.6%

Command-line agentic tasks

In the news

30

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