As it prepares for a potential initial public offering (IPO) by the end of the year, OpenAI is sharpening its focus on its core business, aiming to transform ChatGPT into a highly efficient productivity tool. With ChatGPT’s weekly active users now surpassing 900 million, the company’s primary objective is to convert this massive user base into high-value paying subscribers to compete effectively against rivals like Google and Anthropic. To this end, OpenAI has assembled a dedicated finance team and adjusted its strategy, concentrating resources on its core domains. In terms of financial planning, the company has also lowered its expected computing power investment, now planning to invest approximately $600 billion by 2030 (a reduction from the previously estimated $1.4 trillion). It projects revenues to exceed $280 billion by then, with consumer and enterprise businesses each contributing half. These strategic moves underscore OpenAI’s commitment to steady growth as it approaches its public offering.
On the product front, OpenAI on Tuesday launched its two most capable small models to date – GPT-5.4 mini and GPT-5.4 nano – designed to significantly narrow the performance gap with flagship models while offering lower latency and reduced costs. The GPT-5.4 mini comprehensively surpasses its predecessor in core areas such as coding, reasoning, and multimodal understanding, running more than twice as fast. Its performance on benchmarks like SWE-Bench Pro even approaches that of the larger GPT-5.4 model. Meanwhile, the GPT-5.4 nano is positioned as the lowest-cost, lowest-latency lightweight option, specifically designed for simple programming tasks like data classification and extraction, and is available to developers exclusively via API. The launch of these two models directly addresses the challenge of high latency that has hindered the deployment of large models in real-time interactive scenarios, promising to energize rapidly growing markets such as coding assistants and AI agent systems.
In terms of application, the two models have clearly differentiated roles. The GPT-5.4 mini is available immediately across three platforms: the OpenAI API, the Codex platform, and ChatGPT. The GPT-5.4 nano, however, is focused solely on the API market, offering lower pricing ($0.20 per million input tokens, $1.25 per million output tokens) to serve sub-agent scenarios that require orchestration by more advanced models.
In its announcement, OpenAI emphasized the strategic value of the two new models within multi-model, tiered systems. Using its in-house coding assistant, Codex, as an example, GPT-5.4 handles overall planning and coordination, while sub-agents powered by GPT-5.4 mini execute fine-grained tasks like codebase retrieval in parallel. OpenAI stated that as small models become faster and more powerful, developers can build collaborative systems where large models make decisions and small models execute tasks at scale. GPT-5.4 mini is built precisely for such efficient workflows, representing the most capable small model for this purpose. This architecture is crucial for high-concurrency scenarios like coding assistants and real-time image understanding, allowing developers to strike the optimal balance between speed, cost, and task performance. It enables effective reduction of inference costs without sacrificing the overall intelligence of the system, paving a clearer path for commercial applications.
