Recently, the AI field has been stirred up once again. Sam Altman, co-founder and CEO of OpenAI, revealed in a recent podcast that the highly anticipated GPT-5 is expected to be released this summer. Although the exact release date is yet to be determined, this announcement has already attracted widespread attention within and outside the industry.
GPT-5 Release Imminent, Performance Improvements Generate Excitement
Early feedback from testers indicates that GPT-5 shows significant performance improvements over GPT-4. An ex-OpenAI executive mentioned that if GPT-4’s performance could be likened to an excellent high school student, GPT-5 aims to reach a doctoral level in certain tasks. Looking back at the development of the GPT series, GPT-4 was officially released in March 2023 and achieved significant improvements in multiple areas compared to the previous GPT-3.5. Meanwhile, the GPT-4o released on May 14, 2024, where “o” stands for “omni,” meaning “all-powerful,” further expanded the model’s application scenarios. Currently, GPT-4o has fully replaced GPT-4. As the release of GPT-5 draws closer, the industry believes that the multimodal large model field will enter a new phase of technological competition. Altman also hinted that GPT-5 represents more than just a performance upgrade—it might signify OpenAI’s first real step toward developing a unified, agent-like model. This would bring OpenAI closer to its goal of general artificial intelligence, where the model can not only respond to prompts but also function more like a digital agent, capable of reasoning, planning, and interacting across contexts.
In terms of technical development, OpenAI has invested significant resources into optimizing GPT-5’s underlying architecture. It is reported that the research team has deeply improved the attention mechanism in the Transformer architecture. Traditional full attention mechanisms lead to exponential growth in computational resource usage when processing long texts, severely limiting model performance. OpenAI’s researchers have attempted to introduce a new sparse attention mechanism through a dynamic, hierarchical sparse strategy. This compresses the spatial dimensions of input features at the regional level, allowing fine-grained token-level attention computation in key areas. This hierarchical sparse architecture is expected to improve computational efficiency while more accurately capturing key information in the text, significantly enhancing GPT-5’s ability to handle ultra-long texts, making it more proficient in analyzing academic works, complex legal texts, and other specialized scenarios.
Expanding Cooperation, Tension with Microsoft
In terms of cooperation, OpenAI has been very active. Recently, well-known IP toy manufacturers, such as Barbie, have successfully partnered with OpenAI for its enterprise version of ChatGPT. This has encouraged OpenAI to further expand its B2B customer base. According to foreign media reports, OpenAI has recently launched a discount strategy for its enterprise version of ChatGPT, offering discounts of 10-20% to clients who agree to increase investments in APIs, Deep Research agents, Codex code assistants, and other products. The exact discount rate depends on whether the client signs long-term contracts or bundles multiple services. This business strategy has significantly boosted OpenAI’s enterprise customer base. According to information disclosed by the company to investors, it aims to achieve approximately $15 billion in ChatGPT-related revenue from enterprise customers by 2030. As of earlier this year, ChatGPT Enterprise had already earned $100 million in revenue, and the number of paid enterprise users in its Enterprise, Team, and Edu plans has exceeded 3 million.
However, the relationship between OpenAI and its major supporter, Microsoft, has become increasingly tense. Currently, Microsoft, with its self-developed AI products like Copilot, has entered direct competition with OpenAI. This competition has further escalated since OpenAI’s acquisition of Windsurf. Reports suggest that as the rift between the two companies deepens, some OpenAI executives have started discussing the possibility of Microsoft engaging in anti-competitive practices. Although Microsoft remains OpenAI’s largest investor (holding 20% of shares), OpenAI has also been attempting to reduce its dependence on Azure. The agreement revised in January 2025 allows OpenAI to use other cloud services, provided that Microsoft retains a right of first refusal. Currently, about 30% of OpenAI’s computing power has been migrated to Oracle Cloud and Google Cloud to diversify risks and access more advanced hardware, such as Google TPUs.
Facing Competition, Multi-Dimensional Business Expansion
In response to increasing market competition, in addition to the upcoming release of GPT-5, OpenAI is also making significant moves in business expansion. In terms of developer ecosystem, the GPT-4.1 API, set to launch in April 2025, will support plugin extensions, allowing developers to call external tools (such as real-time search and data analysis). It is expected that by the end of the year, more than 100,000 applications will be integrated. Additionally, ChatGPT Enterprise has introduced a new “custom knowledge base” feature, allowing enterprises to upload internal documents to train custom models. Currently, 2,000 enterprises are paying for this service.
In terms of hardware and infrastructure, OpenAI’s self-owned data center in the Midwest of the United States is scheduled to be operational by the end of 2025, with an initial computing power scale of 100 EFLOPS. This facility will be primarily used for model training and deploying customer-specific instances. The move aims to reduce reliance on third-party cloud services and control costs. Furthermore, OpenAI is actively expanding in global markets. It has partnered with Apple to integrate ChatGPT into iOS 18, expected to cover 2 billion users. It is also launching localized versions in emerging markets such as India and Southeast Asia, supporting languages like Hindi, Indonesian, and 15 others. The goal is to reach 1 billion monthly active users by the end of 2025, a fourfold increase from the current figure.
As the release of GPT-5 approaches, OpenAI’s series of actions in cooperation, competition, and business expansion will undoubtedly deeply impact the future landscape of the AI field. All eyes are on OpenAI to see how it will leverage GPT-5 and its various strategies to continue writing a new chapter in this technological race.
