GPT-5.2 - OpenAI's latest general-purpose AI model series

What is GPT-5.2?

GPT-5.2 is OpenAI's latest advanced artificial intelligence model series, designed for professional work and complex tasks. The model is available in three versions: GPT-5.2 Instant for fast processing of everyday tasks such as information retrieval and translation; GPT-5.2 Thinking optimized for complex tasks such as programming, long document processing, and multi-step project management; and GPT-5.2 Pro focused on high-difficulty tasks, providing high accuracy and reliability, suitable for scientific research and complex analysis. GPT-5.2 has set industry standards in multiple benchmark tests, such as GDPval and SWE-Bench Pro, while reducing the hallucination rate and improving safety and reliability.

Main functions of GPT-5.2

• Professional work optimizationIt excels in creating spreadsheets, building presentations, writing code, handling long documents, and executing complex, multi-step projects, significantly improving work efficiency.
• Strong programming skillsBreaking records in benchmark tests such as SWE-Bench Pro, it can reliably debug code, implement functional requirements, refactor large codebases, and reduce human intervention.
• Long text processingMaintains consistency and accuracy in long document analysis (such as reports, contracts, and academic papers), suitable for in-depth analysis and multi-source workflows.
• Visual understandingIt excels in chart reasoning and software interface understanding, accurately interpreting data dashboards, technical charts, and visual reports.
• Tool callIt enables reliable use of tools in long-chain, multi-round tasks, supports end-to-end workflows, and reduces breaks between steps.
• Science and Mathematics AidIt performs well in benchmark tests such as GPQA Diamond and FrontierMath, and can assist in scientific research and solving complex mathematical problems.
• Security EnhancementIt is more reliable when handling sensitive topics, reduces adverse responses, and protects user security.

GPT-5.2 performance

• Professional knowledge work:
  • In the GDPval benchmark test, GPT-5.2 Thinking achieved a 70.9% win rate, surpassing human expert levels for the first time. It performed exceptionally well in 44 occupational knowledge-based work scenarios, such as creating presentations, spreadsheets, and accounting statements.
  • In internal junior investment banking analyst tasks, the average score for GPT-5.2 Thinking improved by 9.3% compared to GPT-5.1, reaching 68.4%.
• Programming skills:
  • In the SWE-Bench Pro test, GPT-5.2 Thinking achieved a score of 55.6%, setting a new record and significantly outperforming previous versions.
  • In the SWE-bench Verified test, GPT-5.2 Thinking achieved a high score of 80%, demonstrating outstanding performance.
• Long text processingIn the OpenAI MRCRv2 test, GPT-5.2 Thinking achieved near 100% accuracy when processing long documents with up to 256k tokens, making it suitable for deep document analysis.
• Visual understandingIn CharXiv Reasoning and ScreenSpot-Pro tests, GPT-5.2 Thinking showed a significant reduction in error rate and was able to interpret charts, interfaces, and visual reports more accurately.
• Tool callIn the Tau2-bench Telecom test, GPT-5.2 Thinking achieved an accuracy rate of 98.7%, demonstrating excellent performance. Even under the lowest inference intensity, it significantly outperformed GPT-5.1 and GPT-4.1.
• Science and Mathematics:
  • In the GPQA Diamond test, GPT-5.2 Pro achieved a score of 93.2% and GPT-5.2 Thinking achieved a score of 92.4%, demonstrating outstanding performance.
  • In the FrontierMath test, GPT-5.2 Thinking solved 40.3% of the expert-level math problems.
• Accuracy of factsIn anonymized queries using ChatGPT, GPT-5.2 Thinking reduced the error rate by 30% compared to GPT-5.1 Thinking, significantly improving factual accuracy.
• SecurityWhen dealing with sensitive topics such as suicide and mental health, GPT-5.2 showed a significantly lower adverse response rate, demonstrating greater safety and reliability.

How to use GPT-5.2

• Using GPT-5.2 in ChatGPTGPT-5.2 is being rolled out gradually to paid users (including Plus, Pro, Go, Business, and Enterprise plans) starting today, covering Instant, Thinking, and Pro versions. To ensure smoothness and stability, the deployment will be phased, and some users may not be able to use it immediately. During this period, paid users can continue to use GPT-5.1 (legacy mode) for three months, after which GPT-5.1 will be gradually phased out. Users can choose the appropriate version according to their needs.
• Using GPT-5.2 on the API platformIn the OpenAI API platform, all three versions of GPT-5.2 are now fully available, specifically named GPT-5.2 Instant.gpt-5.2-chat-latestGPT-5.2 Thinkinggpt-5.2) and GPT-5.2 Pro (gpt-5.2-proDevelopers can set inference parameters in GPT-5.2 Pro. Both GPT-5.2 Pro and GPT-5.2 Thinking support the new fifth level of inference difficulty. xhighIt is suitable for tasks with extremely high quality requirements. Through the API, users can flexibly integrate the powerful functions of GPT-5.2 to meet the needs of complex application scenarios.

Project address for GPT-5.2

• Project official websitehttps://openai.com/index/introducing-gpt-5-2/
• Technical Papers: https://cdn.openai.com/pdf/3a4153c8-c748-4b71-8e31-aecbde944f8d/oai_5_2_system-card.pdf

GPT-5.2 product pricing

ChatGPT subscription prices remain unchanged.

API pricing

• GPT-5.2 Instant (gpt-5.2-chat-latest):
  • Input: $1.75/million tokens
  • Output: $14/million tokens
• GPT-5.2 Thinking (gpt-5.2):
  • Input: $1.75/million tokens
  • Output: $14/million tokens
• GPT-5.2 Pro (gpt-5.2-pro):
  • Input: $21/million tokens
  • Output: $168/million tokens
• GPT-5.2 offers a 90% discount on cached inputs, meaning: cached inputs: $0.175/million tokens

Application scenarios of GPT-5.2

• Professional document processingGPT-5.2 can efficiently process long documents, such as reports, contracts, and academic papers, supporting in-depth analysis and cross-document information integration, maintaining content coherence and accuracy, and significantly improving work efficiency.
• Programming and Software DevelopmentGenerate, debug, and optimize code in multiple programming languages, support complex front-end development and large codebase refactoring, provide powerful assistance to full-stack engineers, and accelerate the software development process.
• Scientific research and data analysisIt assists in solving advanced mathematical problems and answering complex scientific questions, supports in-depth data analysis and report generation, provides researchers with powerful tools, and promotes academic research progress.
• Customer service and supportThrough multi-round dialogues and tool calls, we handle complex customer issues such as flight delays and lost baggage, providing end-to-end solutions and improving customer satisfaction.
• Content creation and copywritingQuickly generate high-quality copy, news reports, social media content, and creative writing, supporting multiple languages and styles to inspire creativity and meet diverse content needs.