GPT-5: The Epic Update? Meh
Summary:
A plan for the commercial and larger user groups, making OpenAI's consideration more complex.
Phoenix Network Technology
Author | Jiang Fan, Luo Yu
Editor | Dong Yuqing
How many people stayed up all night?
After waiting for over a year, on August 8th at midnight Beijing time, OpenAI officially released the new generation of large language models, GPT-5, and broke through by making the basic version available to all users for free.
OpenAI CEO Sam Altman said in a press release: "If GPT-4 was like a college student, GPT-5 is like a true expert." He described this update as a "technological leap" that's like the first iPhone with a retina display.
Unlike previous versions, GPT-5 is a unified system that no longer separates large-scale models from separate reasoning models. Instead, it consists of three core components: an intelligent and efficient foundation model that can answer most questions; a deep reasoning model (GPT-5 Thinking) for solving more complex problems; and a real-time router (intelligent routing system) that quickly determines which model to use based on dialogue type, complexity, tool requirements, and user prompts.
OpenAI claims that GPT-5 is the strongest AI system yet, achieving significant breakthroughs in programming, mathematics, writing, healthcare, and visual perception.
However, since the press conference just ended, the online community seems to be filled with complaints. Is this really the era of GPT-5?
Will Silicon Valley be jobless?
As usual, it's an old familiar " massacre".
In the latest math reasoning test AIME 2025, GPT-5 scored 94.6% without tools, close to the level of human experts; in the SWE-Bench programming verification test, it scored 74.9%, significantly surpassing the previous model; and in the multi-modal understanding benchmark MMMU, it reached 84.2%, proving that it not only understands text but also deepens its understanding of images, charts, and videos.
Figure | GPT-5 programming score
This new version introduces a real-time router technology that can automatically select the reasoning mode based on task complexity, ensuring depth and reducing computational waste. In simple terms, GPT-5 has learned to "think before acting," achieving both efficiency and accuracy.
Altman also said in the press conference that GPT-5 is capable of writing high-quality software on demand, which will become a hallmark feature of the GPT-5 era.
In OpenAI's testing, this model performed better than all other models in benchmark tests SWE-Bench, SWE-Lancer, and Aider Polyglot. In real-world programming tests, GPT-5 scored 74.9% in SWE-bench Verified and 88% in Aider Polyglot.
During the press conference, Yann Dubois, the person responsible for training work at OpenAI, demonstrated how to use GPT-5 to generate a French learning website with interactive games.
OpenAI's commercialization is accelerating, with paid user numbers increasing from 300 million in June to 500 million currently, and year-over-year revenue (ARR) growing from $100 billion in June to $130 billion currently, expected to exceed $200 billion by the end of the year.
A plan for commercial and larger user groups, making OpenAI's consideration more complex.