A seismic shift is underway in the generative artificial intelligence (AI) market, where OpenAI, maker of ChatGPT, had long run alone. Google's latest AI model, Gemini 3, released last month, is closing in on ChatGPT with significantly improved performance, prompting analyses that the market has entered a two-strong rivalry, with Google and OpenAI fiercely competing for the lead. OpenAI, responding to Gemini's surge, issued a "code red" and released GPT-5.2 ahead of schedule, but the latest model's performance is not overwhelming enough to dominate Gemini, drawing assessments that it falls short of cementing first place.
Industry watchers expect the two companies to continue releasing the latest AI models in quick succession and to announce large-scale investments and infrastructure deals, engaging in a head-on battle to seize leadership of the AI ecosystem.
◇ Despite code red, no performance gap between GPT-5.2 and Gemini 3
As of the 15th, key benchmarks and real-world indicators used to evaluate AI model performance show GPT-5.2 and Gemini 3 at comparable levels. On the LM Arena leaderboard, where users directly rate AI models and rank them, Gemini 3 is No. 1 in an overall assessment spanning text, vision, image editing, search, and more. GPT-5.2 is still early in its release and currently appears on only some LM Arena metrics. Specifically, GPT-5.2 ranked second on LM Arena's WebDev metric, behind Anthropic's Claude Opus 4.5 Thinking. Gemini 3 ranked fourth on that metric. However, Gemini 3 swept first place in detailed metrics such as text, vision, text-to-image conversion, image editing, and search.
Tech outlet Mashable said, "While GPT-5.2's overall LM Arena ranking has yet to be reflected and future results bear watching, so far the two models show almost no difference on key indicators such as benchmark performance and price, and each holds a competitive edge only in specific areas."
Other benchmarks also show GPT-5.2 and Gemini 3 delivering similar performance. On the SWE benchmark, which evaluates software engineering ability including coding, GPT-5.2 scored 80%, surpassing Gemini 3 Pro's 76.2%, and on HLE, a Ph.D.-level reasoning benchmark dubbed the "last exam for humans," GPT-5.2 achieved a 50% accuracy rate with tools, beating Gemini 3 Pro's 45.8%. However, without tools on HLE, GPT-5.2's accuracy was 34.5%, trailing Gemini's 37.5%.
On GPQA Diamond, which measures scientific problem-solving, GPT-5.2 scored 92.4%, slightly ahead of Gemini 3 at 91.9%, while on the MMMLU benchmark, which evaluates multilingual comprehension and advanced reasoning, Gemini 3 (91.8%) outscored GPT-5.2 (89.6%). In addition, on the AIME 2025 (U.S. math competition) benchmark, without tools, GPT-5.2 (100%) held an edge over Gemini 3 (95%).
◇ OpenAI loses performance edge… who will be the final winner?
Shaken by ChatGPT's debut in 2022, Google issued a code red, and three years later has caught up with OpenAI, prompting forecasts that the winner of their AI power struggle will be decided by who first builds a dominant AI ecosystem. Observers say the competitive dynamic between Android and iOS in the past smartphone industry is likely to appear in the generative AI market.
Unveiling GPT-5.2 on the 11th (local time), OpenAI described it as "the most powerful model optimized for professional knowledge work." In fact, GPT-5.2 is being assessed as a "workplace AI" strong in complex, multi-step tasks such as long-document analysis, spreadsheet creation, presentation building, coding, and translation.
However, industry officials say Google holds a competitive edge in accessibility. Gemini 3 is integrated into widely used Google services, including the company's deep search "AI mode," the AI-based summarization tool NotebookLM, Gmail, YouTube, and productivity tools, as part of a strategy to secure users. Moreover, while Gemini includes both image and video generation, ChatGPT supports only image generation and requires a separate Sora app for AI video, a limitation, according to some assessments.
An AI industry official said, "Even if GPT-5.2 later proves to be more capable, considering that Gemini 3 is integrated across Google's ecosystem, the scope of what an individual user can do is broader than when leveraging ChatGPT." Aware of these limits, OpenAI is pushing hard to expand its ecosystem, recently partnering with Adobe to enable Photoshop with ChatGPT and signing a character licensing deal with Walt Disney Company. Disney will invest $1 billion (about 1.47 trillion won) in OpenAI to acquire equity and, in return, will make more than 200 characters, including Marvel, Pixar, and Star Wars, available in ChatGPT and Sora.
Google, which already possesses full-stack capabilities spanning cloud, search, platforms, and AI chip design, also has an advantage in ecosystem expansion, posing a threat to rivals including OpenAI. Gemini can continue training on vast proprietary data from YouTube, Google Search, and Maps, making it easier to improve model performance than for startups like OpenAI or Anthropic, which must secure data while sidestepping copyright disputes. Unlike OpenAI, which relies on external funding, Google's strong finances mean it can continue large-scale infrastructure investments without risking a plunge into deep losses or jeopardizing its business foundation.
OpenAI is expected to focus on regaining its lost technical edge for the time being. Feeling the pressure from Google's fierce pursuit, OpenAI CEO Sam Altman instructed staff to "focus the entire organization's capabilities on improving ChatGPT performance" and signaled that after unveiling another new model early next year, the company will lift the code red. ChatGPT currently has about 900 million weekly users, giving it a commanding No. 1 share in the AI chatbot market, but its growth rate is slowing. According to market research firm Sensor Tower, from August to November this year, ChatGPT's monthly active users (MAU) rose 6%, while Gemini's users surged 30% over the same period, buoyed by the breakout of its AI image generation tool Nano Banana.