‍‍‍On February 12, OpenAI CEO Sam Altman said OpenAI will launch its flagship GPT-5 in the next few months and integrate its AI model into a more unified product.

01.Plan for launching GPT-4.5 and GPT-5

Sam Altman in X The platform released a series of content, disclosed the relevant schedule, and shared other details about the ChatGPT developer product roadmap.

Before the official release of GPT-5, a large language model with a slightly inferior capability, will be launched, which is expected to be in the next few weeks Online within. This model is the successor to the GPT-4o model.

Although GPT-4o is not as good as OpenAI's optimization reasoning ability in tasks such as generating code Language Model (LLM), but it has significant advantages in inference cost.

Altman pointed out in detail that GPT-4.5 will be the last large language model of OpenAI that does not carry the Chain of Thought mechanism.

CoT is a machine learning technology that improves model performance by breaking complex tasks into a series of simple steps. This method has been shown to significantly improve the output quality of LLM, especially when dealing with complex inference tasks. Last year, a source revealed to The Information that the improvement in output quality of GPT-4.5 will be much smaller than the performance of its previous generation GPT-4 when it was released.

In addition, the upcoming new model has not been reported to perform "significantly better" than the GPT-4 in programming tasks. This may be because OpenAI puts more energy into its inference-optimized LLM that is optimized for programming and mathematical tasks.

OpenAI's latest inference-optimized LLM-o3, debuted in December last year. It scored a high score of 25.2% in the "Frontier Math" benchmark, whichOne test included some particularly difficult math problems, with the previous highest score of less than 2%.

It is learned from Altman's X that OpenAI has canceled its plan to release o3 as an independent product. Instead, the company plans to integrate the model into GPT-5, the next generation model after GPT-4.5, which is expected to be launched within a few months.

Altman said: "One of our important goals is to create a tool that can use all the tools." Systems, unified O-series models and GPT series models, these systems can determine when long-term thinking is needed, when not needed, and are widely applicable to various tasks. ”

It is reported that OpenAI plans to offer GPT-5 in both the free and paid versions of ChatGPT. Paid users will be able to access the LLM version with more advanced reasoning capabilities.

In the new GPT-5, voice mode, canvas, search and in-depth research functions will also be built-in, so that chatbots can automatically adapt to users' needs. No need to switch between different models and functions.

These advanced features may be powered by OpenAI's test-time computing technology. This technology improves the quality of model output by investing more time and computing resources to process prompts. OpenAI has applied this technology in o3.

02. New strategy for OpenAI to cope with market competition

Last month, OpenAI also launched A tool called Deep Research, powered by O3, is designed to help users search for information on the web. It is able to generate long reports based on data from hundreds of public sources.

Altman also revealed that free version of ChatGPT users will soon be able to run Deep Research queries twice a month, while ChatGPT Plus subscription users will have up to 10 runs per month permissions for the query. Previously, OpenAI had firmly taken the lead in the AI ​​field, but this position has been challenged recently with DeepSeek's launch of its inference model R1.

The company reportedly developed the model at an extremely low cost and provided chat services for free, while Its API access price is also much lower than OpenAI's model.

In addition to shaking the stock market, the rise of DeepSeek has also allowed OpenAI and other AI companies to re-examine their business strategies, proving that AI models can be lower cost and higher Develop with ease.

Perhaps, Altman's announcement represents a change in strategy.

Unlike previous offering exclusive or early access to certain model features through the Pro program for $200 per month, the new strategy seems to be more inclined to be as wide as possible Promote its products and beat competitors through market share and user experience.

This adjustment may be to deal with increasingly fierce market competition, especially competitors like DeepSeek that sell with low cost and high accessibility. By lowering the threshold and expanding the user base, OpenAI may want to consolidate its market position while attracting more users to rely on its ecosystem.