OpenAI’s GPT-5.2 adds better reasoning, long-context handling and tool integration.
OpenAI CEO Sam Altman called GPT-5.2 “the smartest generally-available model in the world”
Photo Credit: Reuters
OpenAI released the GPT-5.2 family of artificial intelligence (AI) models on Thursday. The second major update to the fifth generation of the GPT model comes with significant performance gains. The company said the AI model, in particular, offers improvements in “economically valuable tasks.” These are essentially productivity tasks, such as sales presentations, accounting spreadsheets, urgent care schedules, and more. The AI giant also claimed that GPT-5.2 outperforms Google's Gemini 3 Pro in several benchmark tests. It will be made available to users via a phased rollout.
In a blog post, the San Francisco-based AI firm announced the latest AI model upgrade and detailed its capabilities. GPT-5.2 AI Models (Instant, Thinking, and Pro) will first be rolled out to the paid subscribers, including ChatGPT Go, Plus, Pro, Business, and Enterprise. The free tier and the Edu accounts are likely to receive the update at a later time. Notably, paid users (except Go) can access GPT-5.1 under legacy models for the next three months. After that, it will be retired.
While the main blog post does not mention Gemini 3 Pro, OpenAI CEO Sam Altman posted a model card of GPT-5.2 to showcase several benchmarks where it fares better than Google's frontier model.
GPT-5.2 benchmark performance
Photo Credit: X/@Sama
One of the central improvements in GPT-5.2 is long-context understanding. Earlier models could remember a certain amount of previous text in a session, for example, the start of a long essay or a business report, but might lose track of details as the text grew. GPT-5.2 is designed to maintain continuity over longer inputs, so it can make sense of larger documents or extended discussions without losing the thread.
Another key area of upgrade is reasoning and task execution. OpenAI says the model is better at handling multi-step tasks, such as building a spreadsheet, composing a presentation, writing or debugging code, or working through a series of logical steps to solve a problem. Essentially the zero-shot responses have been improved, meaning users will have to correct the AI via multiple prompts and iterations less frequently.
GPT-5.2 also brings improvements in agentic tool-calling. This means the model can invoke external functions, apps or tools, such as calling a calendar application programming interface (API) , interacting with a spreadsheet tool, or triggering code execution, more reliably and with better internal checks.
Additionally, the company has also improved the vision capabilities of the model as well. It can now better interpret dashboards, product screenshots, technical diagrams, and visual reports. And during everyday usage, OpenAI claims that the responses of the model in ChatGPT will be more structured, reliable, while retaining the warmer personality.
Apart from ChatGPT, the model is also being released as an API for developers. The base model is priced at $1.75 per million input tokens and $14 per million output tokens. The GPT-5.2 Pro is priced significantly higher at $15 for input and $168 for output (per million tokens).
Catch the latest from the Consumer Electronics Show on Gadgets 360, at our CES 2026 hub.