After sounding the "red alert" in response to competition from Google, OpenAI launched GPT-5.2 on December 11th local time, which includes GPT-5.2 Instant, Thinking, and Pro modes, just one month after OpenAI updated GPT-5.1.
The release of GPT-5.2 is seen by many as OpenAI's counterattack against Google's challenge. Following Google's release of Gemini 3 last month, a redistribution of AI power has been sparked in Silicon Valley, challenging OpenAI's dominance in large-scale modeling.
Unlike GPT-5.1, which emphasized "emotional value" and the ability to engage in pleasant conversations, OpenAI has brought more concrete measures to meet this challenge. GPT-5.2 introduced more intelligent updates and released benchmark scores. Significant improvements can be seen in some benchmark scores.
For example, in the knowledge-based work task GDPval test, the GPT-5.2 Thinking score was 70.9%, significantly exceeding the GPT-5.1 score of 38.8%. In the abstract reasoning ARC-AGI-2 benchmark test, the GPT-5.2 Thinking score was 52.9%, significantly exceeding the GPT-5.1 score of 17.6%. Scores also improved in other benchmark tests. In the Software Engineering SWE-Bench Pro, Science Problems GPQA Diamond, Science Chart Problems CharXiv Reasoning, and Mathematics Competition HMMT tests, the GPT-5.2 Thinking scores were 55.6%, 92.4%, 88.7%, and 99.4%, respectively, while the GPT-5.1 scores were 50.8%, 88.1%, 80.3%, and 96.3%.
Based on these enhanced capabilities, OpenAI claims that GPT-5.2, designed for knowledge-based jobs, is the company's most powerful model to date. "GPT-5.2 has set industry benchmarks in numerous tests. For example, in the GDPval test, this model outperformed industry experts in explicitly defined knowledge-based job tasks covering 44 occupations."
Google's Gemini 3 Pro previously dominated benchmark rankings, but OpenAI has finally turned the tables in this benchmark ranking.
According to previously released data from Google, in the ARC-AGI-2 test, the Gemini 3 Pro scored 31.1%, far exceeding GPT-5.1's 17.6%. In the GPQA Diamond test, the Gemini 3 Pro scored 91.9%, surpassing GPT-5.1's 88.1%. This significant improvement led industry insiders to predict that "it would be difficult for any company to surpass this achievement within the next six months." This time, GPT-5.2 finally outperformed the Gemini 3 Pro in both of the aforementioned benchmark tests. However, it's worth noting that some scores released by Google that significantly exceeded OpenAI's benchmarks, such as the Humanities Last Exam, were not released by GPT-5.2 this time.
OpenAI also emphasized the usability of the new model in professional work, stating that benchmark scores demonstrate GPT-5.2's superior performance compared to professionals in creating presentations and spreadsheets, with significantly improved complexity and formatting compared to the previous generation. However, users will need a paid subscription to use the new spreadsheet and presentation features. Long contextualization capabilities enable the new model to handle documents such as reports, contracts, and research papers. In coding tasks, GPT-5.2 can more reliably debug production code and deliver fixes with less human intervention.
OpenAI demonstrated several coding examples, such as how GPT-5.2 can generate a wave simulator and a holiday greeting card generator with just a prompt. The wave simulator allows users to manipulate values to change wind speed and wave height. OpenAI also highlighted that GPT-5.2 Thinking has a lower illusion rate than its predecessor; in a set of de-identified queries, the new model answered incorrectly 38% less frequently than GPT-5.1 Thinking. OpenAI states that this means the model makes fewer mistakes in writing, research, analysis, and decision-making. GPT-5.2 Thinking also reduces its error rate by approximately half in graph reasoning and software interface understanding. Furthermore, OpenAI claims that GPT-5.2 Pro and GPT-5.2 Thinking are currently the most helpful models for accelerating scientific research.
GPT-5.2 Instant, Thinking, and Pro were launched on ChatGPT on Thursday, with paid subscribers getting early access. However, the "red alert" that OpenAI is raising in response to competition from Google and others has not been lifted with the release of GPT-5.2.
Previously, OpenAI CEO Sam Altman acknowledged in an internal memo that the company was facing the dual challenges of a “tense atmosphere” and “economic headwinds” as competitors such as Google made rapid progress.
OpenAI stated that issuing the "red alert" was a way to focus resources and prioritize tasks, and the company has indeed increased resources related to ChatGPT. Altman indicated that Google's Gemini 3 release had a smaller impact on some of the company's metrics than initially anticipated, but when competitor threats emerge, focus and swift action are crucial. OpenAI expects to end the "red alert" status by January of next year.
GPT-5.2 will not be the only product OpenAI is throwing out in response to competition. Altman said on social media that OpenAI will also be giving away some "little Christmas gifts" next week.
(Article source: CBN)
