OpenAI on Thursday released GPT-5.2, the latest series of artificial intelligence models launched in response to rivals Google and Anthropic, which are quickly catching up in the AI race.
The San Francisco-based AI giant’s flagship GPT‑5.2 is said to bring in significant improvements to the previous model’s general intelligence, long-context understanding, agentic tool-calling, and vision.
"In ChatGPT, GPT‑5.2 Instant, Thinking, and Pro will begin rolling out today, starting with paid plans. In the API, they are available now to all developers," OpenAI limits said.
While Instant is for daily tasks, Thinking is for utilising detailed reasoning for complex work. Pro gives you the highest quality answers for complex and technical problems.
The aim here is to improve its end-to-end processing power for complex, real-world tasks more than the previous models, OpenAI said.
This comes days after OpenAI chief Sam Altman declared a ‘code red’ at the company to fight off competition from ChatGPT’s AI rivals Google Gemini and Anthropic’s Claude, as per a Wall Street Journal report.
A ‘code red’ effectively means OpenAI will pause other product efforts and focus solely on optimising "what (people) want"which is the latest version of ChatGPT.
Business use has also been a major focus in OpenAI’s benchmark tests this year.
In its statement, OpenAI revealed that GPT-5.2 underwent the GDPval, one of the biggest tests that checks how well an AI model—here, the GPT-5.2 Thinking—would fare in 44 real-world careers.
According to the test results, GPT-5.2 Thinking reportedly matched or outperformed industry professionals on 70.9 per cent of tasks, which is almost double the score of GPT-5.
ChatGPT-5 was beaten by Google’s Gemini Pro 3 in handling multimodal tasks and messy data reasoning just weeks ago, at which time the latter was also at the top of the LMArena leaderboard.
The new model also comes with fewer hallucinations, which had been a major advantage of the Gemini Pro 3 over ChatGPT-5.
The new model is also reportedly better at handling complicated tasks with multiple steps that may even involve external tools, OpenAI said, citing GPT-5.2’s 98.7 per cent accuracy in telecom-based customer support scenarios in the Tau2-bench Telecom test.
Comments are closed.