This is the latest ChatGPT

OpenAI has officially launched GPT-5, the latest version of its large language model with many breakthrough performance improvements, along with a series of new features on ChatGPT.

GPT-5 was officially launched on August 8. Photo: Reuters .

On the morning of August 8, OpenAI officially launched GPT-5, the long-awaited version of its large language model (LLM). This is the first major update with a change in version number since GPT-4 was released more than two years ago. Notably, GPT-5 will be available to all users, including free users.

OpenAI CEO Sam Altman likened the move to the iPhone's move to a Retina display, claiming that GPT-5 feels "like talking to a PhD-level expert on any topic."

“Like talking to an expert”

According to CEO Sam Altman, the difference between GPT-5 and previous versions is very clear. This improvement promises to change the way users interact with AI, turning it from a simple assistant into a more in-depth consulting tool.

“I tried going back to GPT-4 and it was pretty miserable. This new model is significantly better in both the obvious and subtle aspects,” he shared.

In this update, OpenAI's inference model is set as the default in GPT-5. This means the system will automatically solve complex problems by breaking them down, resulting in a more logical and insightful answer. Previously, this feature could only be enabled with the "Think Longer" option.

GPT-5’s processing speed is also a highlight. “It’s so fast that sometimes I wonder if it’s really thinking enough. And then it comes up with a great answer,” Altman said.

CEO Sam Altman at the GPT-5 launch. Photo: OpenAI/YouTube.

Notably, OpenAI claims that GPT-5 has a lower “hallucination” rate, providing more accurate answers than any previous system. This helps GPT-5 better understand conversations, documents, or long snippets of code without losing context.

GPT-5 also comes with a new safety feature called “Safe Completions.” Instead of refusing to answer potentially harmful questions, the model will try to provide the most helpful answer within safe limits.

“If someone asks how much energy it takes to burn a particular material, it could be a bad guy or it could be a student asking about physics. With the new feature, the model will only provide information, it cannot be abused to do harm,” explains Alex Beutel, Safety Lead at OpenAI.

OpenAI also emphasized GPT-5's programmability. During the demonstration, Yann Dubois, Lead Model Trainer at OpenAI, asked GPT-5 to create a French learning app. In just a few minutes, the model successfully coded a full web app complete with audio and gaming functionality.

New Features on ChatGPT

With GPT-5, OpenAI has simplified the user experience by implementing an intelligent routing system. Instead of having to choose between models, the interface automatically routes queries to the most appropriate version, depending on the complexity of the question and the user's account level.

Specifically, free users will have access to GPT-5 until their usage limit is reached, after which they will be switched to a more compact version called GPT-5-mini, while Plus package users will have a higher “usage limit”.

The Pro plan provides unlimited access to GPT-5, along with a more powerful version called GPT-5 Pro. Pro users also get GPT-5-thinking, a feature that allows the model to process queries for longer than usual. Developers using the API get access to all three models, including gpt-5-nano, a super lightweight, lower-cost version.

The most notable point is that GPT-5 will be free for all users. Photo: Securityonline.

Along with the launch of GPT-5, the ChatGPT platform also added many new features to enhance user experience.

Pro users can connect ChatGPT to Gmail, Calendar, and Google Contacts starting next week, with other accounts coming soon. OpenAI is also testing four preset personas for the model, including “skeptic,” “robot,” “listener,” and “nerd.”

Users can customize colors for conversations. Advanced Voice is also improved to better understand instructions and adjust speaking style, resulting in more natural and effective interactions.

GPT-5 is “a significant step forward on the path to AGI,” CEO Sam Altman said in a press release. While he stopped short of saying that GPT-5 has achieved artificial general intelligence (AGI), Altman said that the latest version is “clearly a model with general intelligence.”

However, he also added that GPT-5 still lacks some important features to reach AGI. One of the most important reasons is that this model still lacks the ability to continuously learn on its own during operation.

Source: https://znews.vn/gpt-5-chinh-thuc-ra-mat-post1575067.html