TMTPOST -- OpenAI has officially released GPT-5, its new flagship AI model, making it available to all ChatGPT users and developers.The company's CEO Sam Altman characterized GPT-5 as a major step forward in AI development, comparing its debut to a pivotal tech moment: “Something that I just don’t wanna ever have to go back from,” he said, likening the leap to seeing a Retina display on the first iPhone.The company claims GPT-5 is significantly faster, more capable, and less prone to generating inaccurate information. “GPT-3 sort of felt like talking to a high school student,” Altman said during a recent press briefing. “You could ask it a question. Maybe you’d get a right answer, maybe you’d get something crazy. GPT-4 felt like you’re talking to a college student. GPT-5 is the first time that it really feels like talking to a PhD-level expert.”Despite ChatGPT now serving nearly 700 million weekly users, OpenAI has not been at the forefront of large-scale model development in recent months. The release of GPT-5, the company says, changes that. “This is the best model in the world at coding,” Altman said. “This is the best model in the world at writing, the best model in the world at health care, and a long list of things beyond that.”Within ChatGPT, GPT-5 appears as a single unified system rather than offering separate modes for different tasks. A routing mechanism developed by OpenAI automatically activates more advanced reasoning capabilities for complex queries or when prompted with instructions like “think hard.” “The vibes of this model are really good,” said Nick Turley, head of ChatGPT. “I think that people are really going to feel that, especially average people who haven’t been spending their time thinking about models.”GPT-5 is available immediately to all ChatGPT users. However, free-tier users will encounter a limit on how many prompts they can send before the system falls back to a smaller, less powerful “mini” version. For developers accessing GPT-5 through OpenAI’s API, the model is available in three versions—GPT-5, GPT-5 mini, and GPT-5 nano—each offered at different price points.OpenAI is also introducing new customization features for ChatGPT. Users can now choose between four personality styles: “Cynic,” “Robot,” “Listener,” and “Nerd.” In addition, users can personalize the interface by changing chat thread colors.Altman believes GPT-5 will usher in what he describes as “software on demand.” Internally, the model has achieved top scores on several coding benchmarks, including SWE-Bench, SWE-Lancer, and Aider Polyglot.During the press event, Yann Dubois, OpenAI’s head of post-training, demonstrated GPT-5 generating a full website for learning French, complete with an interactive game. In a matter of seconds, the model produced hundreds of lines of code and built the site’s frontend, which Dubois navigated live.According to Alex Beutel, OpenAI’s lead on safety research, GPT-5 underwent over 5,000 hours of evaluation to identify and mitigate risks. A central focus was “making sure the model doesn’t lie to users.” While GPT-5 hallucinates less often than OpenAI’s earlier o3 reasoning model, the tendency to confidently produce false information remains a concern.This issue becomes more serious when the model performs agent-like tasks. Still, OpenAI claims GPT-5 handles multi-step instructions more effectively. “In the past, we’ve seen cases where the model would say it could complete a task that it didn’t actually complete,” said Beutel. “This is a problem.”The new model also introduces what OpenAI calls “safe completions”—a way for GPT-5 to respond more responsibly to prompts that might previously have been blocked outright. “If someone says, ‘How much energy is needed to ignite some specific material?’ that could be an adversary trying to get around the safety protections and cause harm,” Beutel explained. “Or it could be a student asking a science question to understand the physics of this material. This creates a real challenge for what is the best way for the model to reply.”In such cases, Beutel said, GPT-5 “tries to give as helpful an answer as possible, but within the constraints of remaining safe.” The model is designed to respond with high-level, non-actionable information that minimizes potential misuse.Another area of improvement: GPT-5 is better at recognizing when it cannot answer a question or complete a task, a feature OpenAI hopes will foster user trust. However, the company has declined to disclose details about the training data used to develop the model.“This is clearly a model that is generally intelligent.”OpenAI’s long-term mission is to build artificial general intelligence (AGI). Altman believes GPT-5 marks a meaningful step toward that goal, even as the broader field begins to focus on developing “superintelligence.”“I kind of hate the term AGI because everyone at this point uses it to mean a slightly different thing,” Altman said. “But this is a significant step forward towards models that are really capable. This is clearly a model that is generally intelligent.”Still, Altman acknowledged the model has limitations.“This is not a model that continuously learns as it’s deployed from the new things it finds, which is something that, to me, feels like it should be part of AGI.”更多精彩内容，关注钛媒体微信号（ID：taimeiti），或者下载钛媒体App