A few days ago, we shared how Sam Altman’s OpenAI had decided to declare a “red ode“, a way of expressing that they needed to accelerate their efforts in the race with Google and other major players in the field of generative AI in order not to fall behind. That decision involved, among other things, diverting resources and halting secondary projects with the aim of improving the core ChatGPT experience, prioritizing speed, reliability, and personalization. Merely two weeks later, we already have the first release of this new era for OpenAI: GPT 5.2 has arrived.
Indeed, OpenAI is thrilled to discuss the new GPT 5.2, a family of models that “sets a new standard across multiple benchmark evaluations” and, according to their measurements, represents an “improvement in spreadsheet creation, presentation development, code writing, image interpretation, understanding extended contexts, tool usage, and the management of multi-step complex projects.”
In fact, at one point in its presentation, OpenAI states that GPT-5.2 has moved beyond simple “memorization” to enter the realm of true, fluid reasoning. By surpassing the 90% threshold in the ARC-AGI-1 test (a visual and logic assessment that does not rely on language or data learned from the internet) and exceeding 50% in its most challenging version (ARC-AGI-2), they believe the model can now tackle entirely new and abstract problems with competence comparable to that of a human—something that until now was the “holy grail” separating virtual assistants from genuine thinking intelligence: “it is the first model to cross the 90% threshold, improving upon the 87% achieved by o3‑preview last year, while reducing the cost of achieving that performance by approximately 390 times.”
Rest assured, there is still a significant journey ahead before we reach the levels of the coveted Artificial General Intelligence; nonetheless, OpenAI’s statements emphasize that this breakthrough is not merely academic, but rather a transformation in the economic utility of AI. By attaining this level of reasoning while reducing the cost almost 400-fold, they are signaling that expert-level intelligence is now scalable and accessible to all.
Essentially, they are indicating that the tool is now a reliable reasoning engine: it no longer merely “knows facts,” but genuinely “understands” how to solve complex technical or logical problems (such as mathematics or advanced programming) without hallucinating, thus fulfilling one of the fundamental requirements for considering that a machine has achieved Artificial General Intelligence.
At least if you are using one of the paid ChatGPT plans (Plus, Pro, Business, and Enterprise). The company clarifies that “the rollout will be gradual in order to keep the ChatGPT experience as smooth and reliable as possible; if you do not see it immediately, please try again later. GPT‑5.1 will remain available to paid plan users for three months on previous models, after which it will be discontinued.”
In its announcement, OpenAI explains that they have divided the model into three specialized tools to better adapt to your workflow. The central idea is that the daily experience becomes less rigid and much more fluid, offering more pleasant and reliable conversations depending on what you need at any given moment.
Current ChatGPT model tree, including GPT-4o
For everyday and routine usage, they introduce GPT-5.2 Instant. This version is designed to be agile and direct. Its major innovation is that it adopts a much warmer and more human tone, moving away from robotic responses. It is optimized for tasks such as finding information, translating, or quickly drafting technical texts. The most useful aspect of this variant is its ability to synthesize; it gets straight to the point, presenting the key information from the outset instead of beating around the bush, significantly accelerating your workflow.
However, when the task requires reasoning and not just speed, GPT-5.2 Thinking comes into play. This model is designed for deep work and analysis. It is ideal for situations that require structure, such as solving mathematical problems step by step, summarizing very lengthy documents, or planning complex projects. Its strength lies in the fact that it does not merely provide answers, but assists you in thinking things through, offering helpful details and logical order for decision-making.
“GPT‑5.2 Thinking is the best model to date for professional use in real-world environments. In GDPval, an evaluation that measures well-defined knowledge work tasks across 44 occupations, GPT‑5.2 Thinking sets a new record and is our first model to perform at the level of or surpass a human expert,” explain representatives from OpenAI.
Furthermore, data from tests conducted with GPT‑5.2 Thinking indicate that it hallucinates less than GPT‑5.1 Thinking. Thus, in a set of anonymized ChatGPT queries, erroneous responses decreased by 38%.
As for its image interpretation capabilities, GPT‑5.2 Thinking approximately halves error rates in graph reasoning and software interface comprehension—an essential improvement for accurately interpreting dashboards, software screenshots, technical diagrams, and visual reports.
Finally, for the most demanding challenges, they have developed GPT-5.2 Pro. This is the most powerful option, where the utmost priority is quality and precision, even if that means waiting slightly longer for a response. It is the recommended tool for complex programming or difficult questions where you cannot risk errors, as it has demonstrated a dramatic reduction in the mistakes and hallucinations typically seen in previous versions.
Regarding pricing, the company explains that although the ChatGPT subscription price remains the same, on the API, GPT‑5.2 is priced higher per token than GPT‑5.1 because it is a more advanced model. Thus, GPT‑5.2 is priced at $1.75 per one million input tokens and $14 per one million output tokens.
Image: Gemini
Your email address will not be published. Required fields are marked *
Δ