OpenAI launches GPT-5 with PhD-level intelligence, major performance boost
What's the story
OpenAI has announced the launch of its latest flagship model, GPT-5. The new model will be available to all ChatGPT users and developers. Sam Altman, CEO of OpenAI, described GPT-5 as a major improvement over previous models. He likened it to the first iPhone with a Retina display, something you wouldn't want to go back from.
Enhanced capabilities
PhD-level expertise
Altman emphasized the enhanced capabilities of GPT-5, saying it is smarter, faster, and less prone to errors. He compared the conversational experience with previous models: "GPT-3 sort of felt like talking to a high school student," while "GPT-4 felt like you're talking to a college student." With GPT-5, users will feel like they're interacting with a PhD-level expert.
Market position
Aim to reclaim the frontier model crown
Despite ChatGPT's nearly 700 million weekly users, OpenAI hasn't had an industry-leading frontier model in a while. Now, the company hopes that GPT-5 will put it back on top. Altman claimed that this is "the best model in the world at coding," writing, and healthcare, and a long list of things beyond these tasks.
User experience
A single model for all users
Unlike its predecessors, GPT-5 is presented as a single model in ChatGPT, not as a standard model and separate reasoning model. It employs an OpenAI-developed router that automatically switches to a reasoning version for complex queries or when prompted to "think hard." Nick Turley, head of ChatGPT, said, "the vibes of this model are really good," especially for average users who haven't been focusing on models.
Access details
Prompt limit for free users
OpenAI is making GPT-5 available immediately to all ChatGPT users. However, there is an undisclosed limit on prompts for free users, after which the model router will switch to a less powerful "mini" version. For developers accessing GPT-5 via OpenAI's API, the model will be available in three versions at different price points: GPT-5, GPT-5 mini, and GPT-5 nano.
Personalization features
Four new personality themes added to ChatGPT
OpenAI is also adding four personality themes to ChatGPT for customizing responses: "Cynic," "Robot," "Listener," and "Nerd." Users will also have the option to change colors for individual chat threads. Altman believes GPT-5's coding capabilities will usher in an era of "software on demand." In tests, the model outperformed all others at coding on SWE-Bench, SWE-Lancer, and Aider Polyglot benchmarks.
Safety protocols
Addressing confidently lying issue
OpenAI has tested GPT-5 for "over five thousand hours" to understand its safety risks, according to Alex Beutel, the model's safety research lead. The main focus was on "making sure the model doesn't lie to users." While GPT-5 provides fewer hallucinations than OpenAI's o3 reasoning model, confidently lying remains a problem for large language models (LLMs).
Task handling
What are safe completions?
Beutel explained that the problem of confidently lying becomes more complicated when the model starts completing tasks like an agent. However, he said GPT-5 is better at handling multi-step tasks reliably. The model will provide "safe completions" for prompts it would have previously refused to answer. With safe completions, GPT-5 tries to give as helpful an answer as possible while remaining within safety constraints.