ModelBeat
Models/GPT-5.2
All models
O
openai/gpt-5.2

GPT-5.2 is OpenAI’s proprietary AI model, released Dec 11, 2025. It costs $1.75/M input and $14/M output tokens, and handles a 400K-token context window. Its strongest use case is math, where it ranks in the 79th percentile of the models Model Beat tracks.

GPT-5.2 is the latest frontier-grade model in the GPT-5 series, offering stronger agentic and long context perfomance compared to GPT-5.1.

Context
400K
Input
$1.75 / 1M
Output
$14 / 1M
Released
Dec 11, 2025
Modalities
fileimagetexttext
Strongest atMath top 21%Agentic & Tools top 26%Coding top 32%

Benchmarks

Scores on standardized evaluations. Higher is better — and rank shows where GPT-5.2 lands among all models tracked on Model Beat.

70.0
Intelligence Index
Epoch AI
70th percentile of tracked models
68.0
Coding Index
Epoch AI
68th percentile of tracked models
74.0
Agentic Index
Epoch AI
74th percentile of tracked models

Reasoning

8 evals
ARC-AGI86.2%

Abstract visual reasoning

ARC-AGI-252.9%

Harder abstract reasoning

GPQA Diamond91.4%

Graduate-level scientific reasoning

Humanity's Last Exam27.8%

Frontier of human expert knowledge

MMLU-Pro87.4%

Broad expert knowledge

SimpleBench45.8%

Common-sense trick questions

SimpleQA Verified38.9%

Factual accuracy & hallucination

WeirdML72.2%

Novel ML problem-solving

Coding

5 evals
GSO (code optimization)27.4%

Code performance optimization

LiveCodeBench88.9%

Contamination-free coding

SciCode52.1%

Scientific research coding

SWE-bench Verified73.8%

Real GitHub issue resolution

WebDev Arena1480

Human-rated web development

Math

3 evals
AIME 2024/202596.1%

Olympiad-qualifier math

FrontierMath40.7%

Research-level math problems

FrontierMath Tier 418.8%

Hardest research math

Agentic & Tools

5 evals
APEX34.4%

Multi-step agentic tasks

GDPval (win/tie rate)70.9%

Economically valuable work

METR task horizon5.9 h

Autonomous task length

Terminal-Bench64.9%

Command-line agentic tasks

τ²-bench84.8%

Tool-agent-user reliability

In the news

7

Changelog

4
  • GPT-5.2: ARC-AGI-2 dropped from 72.9% to 52.9% (-27%)benchmark · Aug 5, 2026
  • GPT-5.2: ARC-AGI dropped from 94.5% to 86.2% (-9%)benchmark · Aug 5, 2026
  • GPT-5.2: ARC-AGI-2 improved from 52.9% to 72.9% (+38%)benchmark · Aug 5, 2026
  • GPT-5.2: ARC-AGI improved from 86.2% to 94.5% (+10%)benchmark · Aug 5, 2026

Compare GPT-5.2 with

Frequently asked questions

What is GPT-5.2?

GPT-5.2 is the latest frontier-grade model in the GPT-5 series, offering stronger agentic and long context perfomance compared to GPT-5.1.

Who created GPT-5.2?

GPT-5.2 is an AI model developed by OpenAI, part of the GPT family, released Dec 11, 2025.

How much does GPT-5.2 cost?

GPT-5.2's list price from OpenAI is $1.75 per million input tokens and $14 per million output tokens.

What is GPT-5.2's context window?

GPT-5.2 supports a context window of up to 400K tokens.

How does GPT-5.2 perform on benchmarks?

On standardized evaluations tracked by Epoch AI, GPT-5.2 scores 70.9% on GDPval (win/tie rate), 88.9% on LiveCodeBench, 87.4% on MMLU-Pro.

Is GPT-5.2 open source?

GPT-5.2 is a proprietary model (API access), available through its provider's API rather than as a downloadable open-weight model.

Benchmark data from Epoch AI (CC BY) and Artificial Analysis; pricing, specs & descriptions from OpenRouter. Data updated Aug 5, 2026.