“I tried going back to GPT-4 and it was quite miserable,” OpenAI’s CEO Sam Altman spoke to reporters. “This is significantly better in obvious ways and subtle ways.”
Like its predecessor, GPT-5 powers the chatbots, agents and search tools you’re used to using in ChatGPT or through other apps that use OpenAI’s technology. But the company said this version is much smarter, more accurate and faster. Demonstrations showed it quickly creating custom applications with no coding required, and developers said they’ve worked on ways to make sure it provides safer answers to potentially treacherous questions. (Disclosure: Ziff Davis, CNET’s parent company, in April filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)
The new model should be available for everyone on Thursday, including those who use ChatGPT’s free tier. Here’s what to expect.
Watch this: How You Talk to ChatGPT Matters. Here’s Why
A model for everyone (sorta)
GPT-5, unlike some of OpenAI’s incremental releases, will be rolled-out for all users – from those who use it for free via ChatGPT, to those working at companies that pay big enterprise plans. There are differences in how it looks depending on how and whether you pay for it. Here’s the breakdown:
- For free users: After a certain usage limit, you’ll be able to access GPT-5 until a lighter GPT-5 mini model is available.
- For Plus users: Similar as free users, but higher usage limits.
- Pro Users: Unlimited GPT-5 access and access to the more powerful GPT-5 Pro Model. Enterprise/EDU/Team Users: GPT-5 is the default model. It may take until next week to roll it out for everyone.
GPT-5 is actually a combination of two different models. There is a simple but fast LLM, and a robust reasoning model that can handle more complex questions. A routing program determines which model is best suited to handle the prompt.
Even quicker coding skills
OpenAI highlighted the skills and the speed at which the model can write code. This is not just a feature for programmers. It is easier for the model to solve a given problem by creating a tool when it can write a program. Yann Dubois
a post-training leader at OpenAI, demonstrated the model’s coding abilities by asking it to build an app for learning French. It had created a web app with sound and game functions within minutes. Dubois asked it to create 2 different apps by running the same prompt twice through the model. GPT-5’s rapidity in writing code allows you to test multiple times, and then choose the best result. You can also provide feedback for changes until you are satisfied.
“The beauty is that you can iterate super quickly with GPT-5 to make the changes that you want,” Dubois said. Read more: Do not use ChatGPT to do these 11 things
New Safety Features
OpenAI announced this week that GPT-5 had been tweaked to make it safer. Alex Beutel said that the new model will improve user experience by avoiding misleading or inaccurate information.
The system will also respond differently when you ask a question that could be harmful. Beutel said that previous models would refuse to respond to a potentially dangerous question, but GPT-5 instead will try to provide the safest answer. This can be useful when a question sounds innocent (like the chemistry question posed by a student of science) but is actually more sinister. Beutel said. Voices and colors
OpenAI says that if you prefer to speak with your bots rather than typing, you can expect improvements in voice capability. All users, paid or free, will have access to the Advanced Voice mode, with higher usage limits.
You may also change the color of chats. Some options are only available to paid users. You can also customize your chats by changing the personalities. ChatGPT can be set to be supportive and thoughtful, sarcastic, or any combination of these. You can choose from Cynic, Robot Listener, Nerd and more. These options are opt-in and you can change them at any time. ChatGPT can now read your email
ChatGPT is now able to connect to your Google Calendar and Gmail account, so you can ask it about your schedule. You don’t have to – and you may not even want to – depending on your feelings about sharing private information. But you can enable the connectors so that it will automatically pull info from them before you ask it anything. These connectors will be available to Pro users starting next week. Other tiers will gain access after that. But is this the real way to AGI?
Altman said to reporters that the model is “significant step along the path to AGI,” artificial general intelligence. This term is often used to refer models that are as intelligent and capable as humans. Altman said that it is not yet there. One of the biggest reasons is that it doesn’t learn continuously while it is deployed. OpenAI’s stated aim is to develop AGI, although Altman has said that he doesn’t like the term. It’s also got competition. Meta CEO Mark Zuckerberg is recruiting top AI scientists to create “superintelligence.”
Nobody knows if large language models will be the way forward. Three-quarters (75%) of AI experts surveyed in early this year doubted that LLMs could be scaled up to create something with such a high level of intelligence.
