Khaberni - In the massive spread of artificial intelligence applications, comparing major systems like ChatGPT from "OpenAI" and Gemini from "Google" has become complex, especially with the rapid pace of development.
In December 2025, there were speculations about "OpenAI" falling behind in the AI race, before the company turned things around by launching ChatGPT-5.2, which restored its lead in most rankings.
But with the convergence of capabilities of large language models, superficial comparisons based on a single answer to a single question are no longer sufficient or accurate.
The responses are inherently somewhat random, and the conversation style can easily be customized. Therefore, adherence to benchmark tests remains the most objective way to evaluate true performance, according to a report published by "slashgear".
Here are three key benchmarks where ChatGPT shows its superiority over Gemini, based on the latest available results:
Answering complex scientific questions
The first of these is the GPQA Diamond test, designed to measure scientific thinking capabilities at the PhD level in physics, chemistry, and biology.
This test is known for its questions that are "Google" proof, as they cannot be solved through quick searches but require connecting multiple concepts and avoiding incorrect assumptions.
In this test, ChatGPT-5.2 scored 92.4%, slightly ahead of Gemini 3 Pro, which achieved 91.9%.
For comparison, PhD holders are expected to achieve about 65% only, while the average for non-specialists does not exceed 34%, highlighting the high level of both models, with a slight advantage for ChatGPT.
Fixing real-world programming issues from GitHub
The second benchmark is SWE-Bench Pro (Private Dataset), which measures the AI's ability to solve real programming problems taken from actual reports on the GitHub platform.
This test requires understanding an unfamiliar code base, analyzing the problem description, and then providing a practical, implementable solution.
According to the results, ChatGPT-5.2 successfully resolved about 24% of the issues, compared to only 18% for Gemini.
Although these percentages seem modest, this test is considered one of the hardest in its field, while humans still excel at solving 100% of these challenges, confirming that AI is still far from the level of professional software engineers.
Solving abstract visual puzzles
The third criterion is ARC-AGI-2, which is designed to measure abstract thinking and the ability to infer patterns from limited examples, a field where humans traditionally outperform machines.
In this test, ChatGPT-5.2 Pro achieved 54.2%, surpassing most Gemini versions, where Gemini 3 Pro scored only 31.1%, while the higher cost Gemini 3 Deep Think reached 45.1%.
This area is one of the main strengths of ChatGPT, not only compared to Gemini, but most of its other competitors.
Methodology: Why these particular benchmarks?
The results of AI tests are based on rapidly changing versions, hence the focus was on the latest paid models: ChatGPT-5.2 and Gemini 3.
Only three criteria were chosen to represent a wide spectrum of skills, including scientific thinking, solving programming problems, and abstract thinking.
Although there are other tests where Gemini excels, such as some versions of SWE-Bench or the test Humanity’s Last Exam, the focus here was on the areas where ChatGPT clearly excels.
Studies based on personal preference, such as LLMArena, were excluded, although important, noting that Gemini currently tops user preferences there.
These tests indicate that the race in AI is not decided by a single experiment or personal impression, but by precise numbers and standards, and in this round, it seems that ChatGPT holds the advantage in three primary arenas.
