OpenAI has launched GPT-5.2, its most capable model series aimed at improving knowledge work across industries. The release includes three variants for ChatGPT and the API: Instant, Thinking, and Pro, each designed to support complex workflows, long-context tasks, data analysis, coding, and advanced tool use.
The company said GPT-5.2 was engineered to deliver higher output quality, increased reliability, and more efficient performance for business users and developers.
According to OpenAI, GPT-5.2 delivers improved reasoning, stronger factual accuracy, better document comprehension, and expanded vision capabilities. The model outperforms previous versions across a wide range of benchmarks, setting new records on evaluations that simulate real-world professional tasks.
The company reported that ChatGPT Enterprise users already save 40 to 60 minutes a day with AI, while heavy users save more than 10 hours a week. GPT-5.2 is intended to expand those productivity gains by supporting end-to-end execution of multi-step work.
Major Performance Gains Across Reasoning, Coding, and Long Context
GPT-5.2 Thinking achieved state-of-the-art results on the GDPval benchmark, outperforming industry professionals across 44 types of knowledge work. The model also showed improved performance in investment banking-style spreadsheet modeling, producing more polished and structured deliverables with fewer errors.
In software engineering, GPT-5.2 set a new high on SWE-Bench Pro, demonstrating stronger capabilities in debugging, refactoring, and building production-grade features. Early testers reported significant improvements in front-end development, including complex interfaces and 3D UI elements.
Long-context reasoning is another area where GPT-5.2 shows substantial gains. The model performs more accurately on tasks requiring analysis of documents spanning hundreds of thousands of tokens, making it suitable for research, multi-file projects, and deep synthesis work. GPT-5.2 also improves visual understanding, cutting error rates nearly in half on chart interpretation and software interface tasks.
Tool-calling reliability has also improved. GPT-5.2 reached 98.7 percent on Tau2-bench Telecom, supporting multi-step workflows such as customer support resolution, data retrieval, and multi-agent coordination with fewer breakdowns.
Deployment, Pricing, and Safety Enhancements
GPT-5.2 begins rolling out today across paid ChatGPT plans. In the API, GPT-5.2 Thinking and Instant are available immediately, along with the new GPT-5.2 Pro model. OpenAI noted that GPT-5.2 is priced higher than GPT-5.1 but remains below other frontier models, with overall lower cost to achieve similar output quality due to improved token efficiency.
The update introduces additional safety improvements, including stronger responses in conversations involving mental-health distress and new age-based protections supported by an automated age-prediction model. OpenAI said it is continuing to refine refusal behavior, reliability, and content safeguards while developing future versions of its system.
GPT-5.2’s release reflects ongoing collaboration with partners including Microsoft and NVIDIA, whose infrastructure supports large-scale training and deployment. The company said more specialized versions, including a Codex-optimized variant, will arrive in the coming weeks.
The development also aligns with another major initiative, as Disney and OpenAI recently announced a three-year licensing partnership enabling Sora to generate fan-inspired short videos using more than 200 Disney, Marvel, Pixar, and Star Wars characters.
The announcement follows another milestone for OpenAI, with ChatGPT topping Apple’s 2025 list of most-downloaded free iPhone apps, surpassing TikTok, Instagram, and Google Maps.
