OpenAI has taken the race for artificial intelligence a step further with the unveiling of GPT-5.2, its most advanced model to date for professional and enterprise work . The update arrives at an unusually rapid pace: it appears less than a month after GPT-5.1 and amidst a technological battle with Google, Anthropic, and other competitors who are also accelerating their development paths.
The company maintains that this version is not a complete generational leap, but rather an evolution within the GPT-5 series focused on reasoning, error reduction, and tool coordination. The launch directly impacts ChatGPT users in Spain and Europe, where the assistant's use has become well-established in both educational settings and businesses that integrate it into their daily workflows.
A launch at the pace of "code red"
The announcement of GPT-5.2 comes amid a highly competitive environment. Just days before, Sam Altman internally declared a " code red " to focus resources on improving ChatGPT and temporarily shelve more experimental projects, such as the introduction of announcements or the mass deployment of autonomous agents. The priority, as he conveyed to the staff, was to visibly improve the quality of responses and performance in professional tasks.
OpenAI's applications team has clarified that GPT-5.2 wasn't released early because of that "red code ," but rather had been in development for months. Even so, the company itself admits that the additional teams dedicated specifically to ChatGPT were helpful in refining details and finalizing the version much faster than GPT-5.1.
The acceleration is also explained by the pressure from Gemini 3 , Google's model that has gained ground in recent weeks with users and in various performance tests, especially in general reasoning. In parallel, Anthropic has been pushing in the programming arena with its code-centric models. In this scenario, GPT-5.2 is presented as OpenAI's response to regain ground in the professional-grade AI segment.
At the market level, the company acknowledges that it has reoriented its strategy: less focus on commercial experiments such as advertising and more on strengthening the technological base of ChatGPT and its API to remain relevant to business clients and public administrations, also in Europe.
An evolution within the GPT-5 series
OpenAI presents GPT-5.2 as the next direct iteration of GPT-5.1 , not as a next-generation model. The main goal is to improve knowledge work: programming, document analysis, vision, spreadsheets, presentations, and projects composed of many interconnected steps.
Among the most notable changes is improved handling of long contexts , crucial for companies that manage lengthy contracts, technical reports, or complex legal documentation. The model is able to follow the thread of conversations and larger documents, reducing losses of coherence in later sections.
The company also states that GPT-5.2 better coordinates the use of external tools (APIs, databases, code interpreters, internal systems), a key point for developers and IT departments building agents capable of performing real-world actions beyond simply responding with text. This improvement strengthens the integration of third-party applications into automated workflows.
Although the overall architecture remains consistent with GPT-5, OpenAI claims to have adjusted the training and post-training processes to reduce hallucinations and factual errors . In internal tests compared to GPT-5.1, the Thinking reasoning variant cuts down by approximately one-third the clearly incorrect responses to objective questions.
Three variants: Instant, Thinking, and Pro
As with recent releases, GPT-5.2 comes in three versions designed for different uses: Instant, Thinking, and Pro . Rather than adding exclusive features, the update strengthens the profile of each version to better distinguish what each model is recommended for.
GPT-5.2 Instant is the fastest and most economical option. It's geared towards everyday queries, quick writing, information searches, and study support . OpenAI indicates that it offers more stable answers than its predecessor in this range, with clearer explanations and fewer changes in criteria for similar questions.
GPT-5.2 Thinking is the deep reasoning model and the one that has evolved the most in this generation. It has been trained to handle complex, multi-step tasks , such as advanced programming, financial analysis, interpreting long documents, and detailed project planning. Most of the improvements in mathematics, science, and tool coordination are concentrated here.
Finally, the GPT-5.2 Pro sits at the top of the range: it's the most powerful and expensive model to run , designed for environments where precision is paramount over latency, such as technical research, complex legal tasks, or demanding data analysis projects. It offers the highest level of detail and control, in exchange for greater computing power consumption.
This segmentation aims to allow both individual paying users and businesses and developers to choose the model that best fits their use case , rather than indiscriminately opting for the most expensive one. For European organizations with budgetary or regulatory constraints, this differentiation can facilitate cost and workload planning.
Test results: from desktop to benchmarks
OpenAI backs up the launch with a battery of performance figures. In the GDPval test suite, designed by the company itself to simulate 44 real-world occupations, GPT-5.2 Thinking manages to win or tie against human professionals in around 70,9% of the tasks, completing them much faster and at a significantly lower operating cost.
In the scientific field, the model improves its scores in GPQA Diamond , an assessment that measures high-level reasoning in science, and in FrontierMath , which focuses on advanced mathematical problems. In some test variants, such as AIME 2025, OpenAI reports that GPT-5.2 achieves scores close to 100% on the items considered.
Beyond benchmarks, the company highlights internal evaluations of typical financial analyst tasks , such as three-state models or leveraged buy simulations. In these scenarios, GPT-5.2 Thinking reportedly improved from average scores of around 59% to over 68%, approaching the level of specialized teams.
Improvements are also noted in the generation of spreadsheets and presentations with clearer structures and fewer consistency errors. Technology companies working with OpenAI—such as productivity platforms or European software startups—have reportedly seen advancements in long-range reasoning and the consistent use of automated tools within the model.
The company insists, however, that benchmarks are only part of the story and that real-world performance with end users in Spain and Europe will depend on how these capabilities are integrated into specific use cases: from freelancers preparing budgets or reports, to large companies deploying internal assistants for their staff.
Fewer mistakes and more stability in day-to-day operations
One of OpenAI's core messages is the reduction of hallucinations and factual errors . According to their data, GPT-5.2 Thinking produces 38% fewer clearly incorrect answers than GPT-5.1 in internal tests of objective questions. This is especially relevant for those using the model to support reporting, research, or business decisions.
Beyond accuracy, the company claims to have improved the model's ability to detect its own errors before delivering the final response. In other words, GPT-5.2 performs more internal checks and discards solutions it deems unreliable, resulting in a somewhat more predictable experience during long conversations.
This update comes after controversial incidents in which OpenAI was criticized for the chatbot's role in sensitive conversations , including the case of a teenager who committed suicide after using it as a confidant. Without going into legal details, the company states that GPT-5.2 brings improvements to the handling of sensitive topics, with particular attention to signs of emotional distress.
In practice, the model should better adapt its tone and recommendations when it detects that it is dealing with vulnerable users, although OpenAI insists that it still does not replace professional medical or psychological support, a relevant warning in light of European regulation on AI.
A more comfortable environment for developers and businesses
GPT-5.2 is also directly aimed at those building products and services on the OpenAI API. This reasoning variant demonstrates superior performance in software development tests that require applying comprehensive and consistent changes to real-world projects, not just solving isolated exercises.
Collaborative development platforms and assisted programming tools that already worked with GPT-5.1 report that the new model better coordinates sequences of complex steps , resulting in fewer manual iterations to complete tasks such as refactoring code, applying revisions, or generating modules from textual specifications.
In European business contexts, where integration with internal systems is often more conservative due to regulatory compliance issues, improvements in consistency and change control can facilitate the adoption of agents that perform automated actions with less direct human supervision.
Another relevant aspect is the model's performance with extensive corporate documentation , as GPT-5.2 improves the reading, summarizing, and comparison of contracts, internal policies, or technical reports, something of particular interest to law firms, consultancies, and compliance departments in Spain and the EU.
Availability, deployment, and pricing
OpenAI has begun the global rollout of GPT-5.2 in ChatGPT for its paid plans , including Plus, Pro, Go, Business, and Enterprise. Users of these plans will gradually see the option to select the Instant, Thinking, and Pro variants, while the free version will continue to use previous models.
In the API, GPT-5.2 Thinking is exposed under the name gpt-5.2 , while the faster, chat-oriented version appears as gpt-5.2-chat-latest. OpenAI will keep GPT-5.1 available for a transition period of several months, but its stated intention is to gradually migrate the bulk of usage to the new series.
In terms of pricing, GPT-5.2 charges $1,75 per million input tokens and $14 per million output tokens for the API, higher than the fees for GPT-5.1. Despite this, the company argues that the model's greater efficiency in complex tasks can reduce the overall cost per project by requiring fewer calls and revisions.
For Spanish and European companies that primarily value reliability in analysis, documentation generation, or process automation, OpenAI's logic is to sell GPT-5.2 as an investment in accuracy rather than as a low-cost tool. This positioning aligns with its strategy of targeting the high-value-added corporate segment.
In terms of adoption, ChatGPT remains one of the world's most widely used AI chatbots, with hundreds of millions of weekly active users . The company is confident that improving the daily user experience, making it more structured and predictable, will further boost professional use, especially in markets where it already has a strong presence, such as Spain.
Competition, infrastructure and roadmap
The launch of GPT-5.2 comes at a time when the update cycles for leading models have shrunk from years to weeks . This speed makes it clear that technological advantage is becoming increasingly fleeting and that companies in the sector are incurring rising costs to keep pace.
With Gemini 3, Google has leveraged its privileged access to data and its productivity ecosystem (Docs, Sheets, Drive, Workspace) to integrate AI into tools already ubiquitous in offices worldwide. Anthropic, meanwhile, has focused on security and performance in programming, strengthening its position in specific segments.
To sustain this momentum, OpenAI is strengthening its infrastructure through agreements with chip manufacturers and cloud providers , as well as massive investments in data centers. The company acknowledges that the cost of operating advanced reasoning models has skyrocketed, exceeding the support initially provided by cloud credits from its partners, and cites an agreement with AWS to support its infrastructure as one of its key moves.
At the same time, OpenAI has relied on strategic alliances with large corporations , such as Disney and Walmart , which have seen its models as a way to create new content formats and digital experiences. These agreements, while not directly part of GPT-5.2's functionality, illustrate the company's focus on uses that generate recurring revenue and justify significant infrastructure investments.
In Europe, this expansion will have to coexist with a stricter regulatory framework, marked by the EU Artificial Intelligence Regulation , which will require greater transparency in the operation of models such as GPT-5.2 and in the use that companies and public administrations make of these systems in their services.
With GPT-5.2, OpenAI aims to solidify its position as a leading provider of generative AI for professionals and organizations, strengthening reasoning, reliability, and integration capabilities without introducing a groundbreaking innovation . This move comes at a time when the market is beginning to demand fewer spectacular demonstrations and more stable tools that solve real-world work problems, particularly within the Spanish and European business sectors, where AI adoption is becoming a key driver of competitiveness.
