The new language model GPT-5.2 is now available for paying OpenAI users, in three versions: Instant, Thinking y Pro. He wants to regain the throne of artificial intelligence, which he lost with Gemini 3.
GPT-5.2 Instant It is used for quick and simple queries, such as searching for information, resolving a question or translating.
GPT-5.2 Thinking It is the reasoning model. When the AI has to analyze a response. It is ideal for math tasks, programming, document analysis, or creating plans.
Finally, GPT-5.2 Pro It is the most accurate and reliable version to solve complex problems.
GPT-5.2 improves in everything on its predecessor
OpenAI ensures that GPT-5.2 is more accurate and less mind-blowing than GPT-5.1. “GPT-5.2 is better at creating spreadsheets, creating presentations, writing code, perceiving images, understanding long contexts, using tools, and managing complex, multi-step projects,” the press release says.
This new version is very important for OpenAI, after the code red that Sam Altman launched, because for the first time Gemini 3 surpassed GPT-5.1. At the moment we do not know if it has regained its leadership, because in the benchmarks it has presented it only compares it with its predecessor, not with the competition.
While GPT-5.0 laid the foundations for a unified model that independently chooses the AI used, depending on the task, GPT-5.1 focused on improving conversations, as well as agentic tools, and code creation. Already consolidated, version 5.2 strives to improve everything.
The biggest leap that GPT-5.2 presents is in its reasoning model. According to TechCrunchwho attended a presentation to the press where they did show benchmarks, GPT-5.2 Thinking outperforms Anthropic’s Gemini 3 and Claude Opus 4.5 in almost all reasoning testsfrom real-world software engineering tasks (SWE-Bench Pro) and doctoral-level scientific knowledge (GPQA Diamond) to abstract reasoning and pattern discovery (ARC-AGI suites).
Furthermore, this reasoning model makes 38% fewer errors than the previous version.
GPT-5.2 seems capable of competing head-to-head with Gemini 3, except for one thing: imagingwhich Nano Banana has revolutionized. OpenAI acknowledges that there is nothing new here, but they are already working on a new model for generating images, which they will present in January.