openai/gpt-5.2

The best model for coding and agentic tasks across industries

43 runs

Readme

GPT-5.2

GPT-5.2 is the most advanced frontier model for professional work and long-running agents. It excels at knowledge work, coding, long-context understanding, vision, and tool usage.


Overview

  • Professional Knowledge Work: GPT-5.2 saves 40–60 minutes/day for average users and over 10 hours/week for heavy users.
  • Capabilities:
  • Spreadsheets & presentations
  • Coding & software engineering
  • Image perception & chart reasoning
  • Multi-step reasoning & tool usage
  • Long-context comprehension (up to 256k tokens)

  • Benchmarks & Performance:

  • GDPval (Knowledge Work, 44 occupations): 70.9% wins/ties vs professionals
  • SWE-Bench Pro (Software Engineering): 55.6%
  • GPQA Diamond (Science Q&A): 92.4%
  • FrontierMath Tier 1–3: 40.3%
  • ARC-AGI-2 (Abstract Reasoning): 52.9%
  • CharXiv Reasoning (Scientific Figures): 88.7%

Key Improvements

Coding

  • Stronger front-end & full-stack development
  • Interactive coding, debugging, code reviews, and bug finding
  • SWE-Bench Verified: 80.0%

Long-Context Reasoning

  • OpenAI MRCRv2 (up to 256k tokens): Maintains coherence & accuracy
  • Supports multi-file projects, research papers, and contracts

Vision

  • Improved chart reasoning and GUI understanding
  • ScreenSpot-Pro accuracy: 86.3%
  • Identifies components in images with bounding boxes

Tool Usage

  • Tau2-bench Telecom: 98.7%
  • Handles multi-step workflows like customer support or multi-agent tasks

Science & Math

  • GPQA Diamond: 92.4%
  • AIME 2025: 100%
  • FrontierMath Tier 1–3: 40.3%

Factuality & Safety

  • Hallucinates 30% less than GPT-5.1
  • Improved responses in sensitive conversations, mental health, self-harm, and emotional reliance

Model Variants

Variant Description
GPT-5.2 Instant Fast, capable for everyday work & learning
GPT-5.2 Thinking Deep work, coding, long-context analysis
GPT-5.2 Pro Highest accuracy, complex reasoning, most trustworthy

Availability & Pricing

  • ChatGPT: Paid plans (Plus, Pro, Go, Business, Enterprise)
  • API:
  • GPT-5.2-chat-latest (Instant)
  • GPT-5.2 (Thinking)
  • GPT-5.2-pro (Pro)

Pricing per million tokens:

Model Input Cached Input Output
gpt-5.2 / gpt-5.2-chat-latest $1.75 $0.175 $14
gpt-5.2-pro $21 - $168
gpt-5.1 / gpt-5.1-chat-latest $1.25 $0.125 $10
gpt-5-pro $15 - $120

Partners

  • NVIDIA & Microsoft
  • Azure data centers & NVIDIA H100/H200/GB200 GPUs

Notes

  • GPT-5.2 sets a new state of the art in professional tasks, coding, reasoning, long-context understanding, vision, and tool usage.
  • For critical tasks, human oversight is recommended.
  • GPT-5.1 remains available for three months under legacy plans.

References & Benchmarks

Professional: GDPval, Investment Banking Spreadsheets
Coding: SWE-Bench Pro & Verified, SWE-Lancer
Long Context: OpenAI MRCRv2, BrowseComp, GraphWalks
Vision: CharXiv, MMMU, ScreenSpot-Pro
Tool Usage: Tau2-bench, BrowseComp, Scale MCP-Atlas, Toolathlon
Academic: GPQA Diamond, HLE, MMMLU, AIME, FrontierMath
Abstract Reasoning: ARC-AGI-1 & 2

Model created