Recently, Moon Byung-ro, a professor at Seoul National University’s Department of Computer Science and Engineering, conducted a “simple test” on major artificial intelligence (AI) models. He had the top models of each AI solve the midterm exam questions he created. In the test results, Google’s “Gemini 2.5 Pro” scored first with 74 points. OpenAI’s “ChatGPT,” xAI’s “Groq” developed by Elon Musk, and Anthropic’s “Claude” scored 41, 42, and 15 points, respectively.
The score of Gemini 2.5 Pro corresponded to 6th place even when compared to actual university students taking the “Computer Algorithms” class at Seoul National University. Professor Moon noted, “It is difficult to definitively assess performance based solely on this test result due to the varying strengths of each AI model,” but added, “It seems clear that Google’s Gemini has significantly improved its performance.”
Despite having first proposed the “transformer structure” in 2017, which became the basis for Generative AI, and facing multiple setbacks, Google has launched aggressive moves in the AI sector this year. The signal flare was the “Google Cloud Next 2025” held from April 9-11 in Las Vegas. At this event, Google unveiled a plethora of new technologies, including AI chips, networks, and multi-agent platforms.
The performance of the AI model “Gemini 2.5 Pro” is spreading by word of mouth, causing tension for OpenAI, which has maintained its market lead. Google plans to unveil new technologies related to Android and AI at the annual developer conference “Google I/O 2025” in May.
Two years in the trenches: Google returns
Until last year, Google had been evaluated as lagging behind in the AI leadership race. Google’s AI chatbot “Bard” had to change its name to “Gemini” after providing incorrect answers at its first presentation in 2023. In May 2024, OpenAI unveiled “GPT-4o” a day before Google I/O, causing media attention to inevitably focus on OpenAI’s announcement.
This year, a different trend is sensed. At “Google Cloud Next 2025,” core figures, including Google CEO Sundar Pichai, Chief Scientist Jeff Dean, and Vint Cerf, Google’s Chief Internet Evangelist, all turned out to successfully assert their dominance.
They introduced a range of products, such as the 7th generation Tensor Processing Unit (TPU) “Ironwood,” the global private network “Cloud Wide Area Network (Cloud WAN),” the corporate platform “Vertex AI,” the agent communication protocol “Agent to Agent (A2A),” and the latest AI model “Gemini 2.5 Flash.” It was a large-scale offensive targeting the AI market.
Gemini 2.5 Pro, announced before the event, showcases excellent reasoning ability through the “thinking” process, excelling in complex math, science, and coding problems. The token context window of this model amounts to 1 million tokens. The competing model, OpenAI’s GPT-4o Turbo, offers a context window of 128,000 tokens, while Anthropic’s Claude 3 offers 200,000 tokens. The larger the context window, the better the ability to analyze extensive information such as long papers, codes, and dialogue lists without losing context.
Google has also aggressively priced the API (Application Programming Interface) fees. For Gemini 2.5 Pro, the input fee is set at $1.25 per 1 million tokens (approx. 1,770 won for up to 200,000 tokens), and the output fee is set at $10 (approx. 14,000 won). If it exceeds 200,000 tokens, the input fee is $2.50 (approx. 3,500 won), and the output fee is $15 (approx. 21,000 won).
In the developer community, there is an evaluation that “Google does not slow down AI development to protect its search market” and “Google’s complex organization has been overhauled to focus on AI.” Kim Hak-hoon, CEO of Knowledge Cube, which has a cooperative relationship with Google, predicted, “Google aims to transition from an ad-centric business to a subscription-centric business in the long term.”
Recently, the U.S. IT specialist magazine “Wired” reported that “there has been a two-year struggle by Google to catch up with OpenAI.” According to the report, Google’s board demanded real-time reports from the AI organization and integrated the previously complex AI organization (DeepMind and Google Brain). CEO Pichai executed mass layoffs to increase organizational tension, which ultimately extended developer working hours.
A tense OpenAI, “We will acquire Chrome”
Faced with Google’s intense pursuit, OpenAI is on high alert. Recently, OpenAI launched the next-generation flagship model “GPT-4.1” and reduced API prices by up to 75%. OpenAI, which led the AI market until 2024, finds itself in a position of responding to Google’s offensive. The pricing pressure triggered by Google is expected to lead to a competitive bleeding among AI model developers for some time.
There is also an analysis suggesting that OpenAI’s recent allowance of trailing competitors is not unrelated to OpenAI’s leadership crisis last year. Last year, key figures like Chief Scientist Ilya Sutskever and “Mother of ChatGPT” Mira Murati, the Chief Technology Officer (CTO), left en masse. Currently, only 3 out of the 11 founding members of OpenAI remain.
However, the number of ChatGPT users is OpenAI’s greatest weapon. With the success of the Studio Ghibli-style image generation feature, ChatGPT’s weekly active users (WAU) surpassed 500 million. It increased by more than 30% in about three months from around 350 million by the end of 2024. Hwang Gyu-jong, CEO of Waker, said, “The first-mover advantage enjoyed by OpenAI is larger than expected,” adding that “The vast query and response data generated by users is a crucial asset that accelerates the learning and improvement of AI models.”
Recently, OpenAI unveiled the “Memory with Search” feature. It offers customized web search results based on information obtained from past user conversations. This personalized search function can enhance the customer lock-in effect. A contributor to the U.S. economic media “Forbes” pointed out, “ChatGPT captures writing style, project history, and work habits in just six months,” adding, “For users, moving to another platform incurs significant transition expenses.”
Recently, Google decided to provide Gemini Live’s camera and screen sharing features for free to all Android users to establish a virtuous cycle of “user acquisition → data acquisition → quality improvement.”
Amid this, antitrust lawsuits facing Google are emerging as another variable in the AI feud. The U.S. Department of Justice is demanding strong antitrust measures, including the sale of Google’s internet browser “Chrome.” A U.S. court is expected to decide on Google’s illegal monopoly resolution plan in the search market by August, based on the proposals from the Department of Justice. Previously, on April 17, Google lost its antitrust lawsuit in the online advertising technology market.
OpenAI expressed interest in acquiring the Google Chrome browser in the event of a forced sale order. Nick Turley, product head of OpenAI’s ChatGPT, testified as a witness during the remedies hearing of Google’s antitrust lawsuit held in Washington, D.C., on April 22, stating, “Like other corporations, we are interested in acquiring the Google Chrome browser.”