OpenAI unveils GPT-5.2, its new model built for professional work

OpenAI has introduced GPT-5.2, the company’s latest large language model and its most direct response yet to Google’s fast-rising Gemini 3 Pro. Released on the same day OpenAI announced a licensing deal with Disney for Sora, the new model is positioned as a major upgrade for people who rely on AI for real-world tasks and daily productivity.

According to OpenAI, GPT-5.2 is more capable at generating spreadsheets, building presentations, writing and debugging code, working with images, and handling longer, more complex instructions. The company says its internal tests show improvements across 10 major benchmarks.

Stronger reasoning, fewer mistakes

The highest-end version, GPT-5.2 Thinking, delivered some of the largest jumps. On AIME 2025, a benchmark involving 30 challenging math problems, the model scored a perfect 100 percent, surpassing GPT-5.1’s 94 percent result and doing so without using external tools like web search. It also posted a more than 10-point leap on ARC-AGI-1, a test designed to measure abstract reasoning.

OpenAI also claims GPT-5.2 Thinking produces factual errors 30 percent less often, a change aimed directly at users who depend on AI for research, analysis, drafting and decision support. “Fewer mistakes mean a more dependable tool for everyday knowledge work,” the company said.

OpenAI to Allow Mature Content in ChatGPT for Verified Adults

While most users will interact with GPT-5.2 Instant, OpenAI says this faster version has been upgraded as well, especially in everyday tasks like answering info-seeking questions, handling step-by-step tutorials, doing technical writing and providing translations. It builds on the friendlier conversational style introduced in GPT-5.1 Instant.

A high-stakes release after months of criticism

GPT-5.2 arrives at a critical moment for OpenAI. Earlier in the year, GPT-5 launched to mixed reactions, with users describing the model as oddly dull and prone to simple mistakes. The backlash grew loud enough that many people asked the company to restore GPT-4o.

The pressure increased when Google released Gemini 3 Pro, which quickly climbed to the top of LMArena , the community-run leaderboard where people vote on the best model outputs. Reports claim this prompted a “code red” directive inside OpenAI to accelerate improvements.

Before today’s release, OpenAI’s GPT-5.1 was ranked sixth on LMArena, behind models from Anthropic and xAI.

Now, as GPT-5.2 begins rolling out, OpenAI believes the new system can rival Google’s best. The company says GPT-5.1 will remain available for at least three months under a legacy section for paid users who still prefer it.

OpenAI is releasing GPT-5.2 in three tiers, Instant, Thinking, and Pro, all initially limited to users on paid plans.