On March 6, OpenAI launched GPT-5.4, the latest generation AI model. OpenAI defines this new model as the company's "most realistic and efficient" model currently - it consumes less tokens and runs faster. And in ChatGPT, GPT-5.4 thinking improves deep web research. It can better maintain context when thinking for a longer time, and users can interrupt the model, add instructions or adjust the direction mid-answer.
Next, OpenAI will launch GPT-5.4 Pro in ChatGPT and API, targeting users who want the highest performance on complex tasks.
According to disclosures, GPT-5.4 brings together the latest progress made by OpenAI in reasoning, coding, and agent workflows, integrating the industry-leading GPT-5.3-Codex coding capabilities, while improving the performance of the model in various tools, software environments, and professional tasks involving spreadsheets, presentations, and documents. The result is a model that does complex real-world work accurately, effectively, and efficiently—delivering what users ask for with less back-and-forth communication.
In addition, in ChatGPT, the thinking model of GPT-5.4 can already provide its thinking plan in advance, and users can adjust the direction at any time during the answer process. Moreover, GPT-5.4's thinking ability has also been improved, and it is especially good at deep network research, especially when dealing with highly specialized queries; at the same time, it can better maintain contextual coherence to deal with problems that require longer thinking. Together, these improvements result in higher-quality, more responsive answers that are always relevant to the task at hand.
It is worth noting that, combined with the above capabilities, GPT-5.4 has finally become the first model within OpenAI to have Computer Use (computer control) capability built-in natively in a general model. Previous OpenAI related capabilities, such as ChatGPT Atlas, AutoGPT Agent, etc., were plug-in/plug-in/independent agent modes and were not native capabilities of the model.
Before GPT-5.4, products such as Anthropic Claude Opus 4.6 and Google Gemini Advanced had already occupied the minds of developers in the Computer Use field.
Although OpenAI emphasizes that the new model consumes less tokens and runs faster, the result of improved performance is an increase in the overall price of the model. According to official disclosures, the input price of GPT-5.4 has increased to US$2.5/million tokens, and the output price has reached US$15/million tokens. The input price of the Pro version is as high as $30/million tokens, and the output price is $180/million tokens.
Regarding the price increase, some netizens believe that the price only needs to meet the user's ROI. Some netizens reported that they spent US$80 just to say hello to GPT-5.4. "Everyone is saying that GPT-5.4 has reached the level of AGI (Artificial General Intelligence) and is the smartest model, but do you want to ask questions at the AGI level?" The user raised his own question.
Behind the performance improvement and price increase of GPT-5.4 is OpenAI’s revenue pressure in the current AGI competition landscape. Although the company has just announced the completion of US$110 billion in new financing, Nvidia CEO Jensen Huang, an important investor in OpenAI, said that this US$30 billion investment may be the last one, and the US$100 billion supercomputing facility agreement reached by the two parties last year is unlikely to be realized. The core reason is that OpenAI is about to go public, and the capital structure and financing strategy will change, making it difficult for Nvidia to continue to make large-scale capital injections as a private equity investor.
There is also news that this high-value financing of US$110 billion is not a one-time payment of "real money", but a direct agreement with attached conditions. Superimposed on the current pursuit of AI by companies such as Anthropic and Google, OpenAI has adjusted its cost structure to be more pragmatic while expanding its revenue scale through multiple channels, including the recent "cutting off" of the U.S. Department of War order that Anthropic just lost.
The price increase of GPT-5.4 is a test of OpenAI on the eve of its IPO. It is trying to prove to investors and the capital market that OpenAI’s product performance is so high that even if the price is increased, there are still companies and users willing to pay for it.