OpenAI Releases GPT-5 For All Users: Sam Altman Says it is the 'Best Model in the World at Coding and Writing'

OpenAI has unveiled GPT-5, its latest flagship model, now accessible to all ChatGPT users and developers. CEO Sam Altman describes it as a significant advancement from previous models. He likens it to the first iPhone with a Retina display, something he wouldn't want to revert from. The new model is touted as smarter, faster, and less prone to errors.

Despite ChatGPT's impressive user base of nearly 700 million weekly users, OpenAI had not led the industry with a frontier model for some time. Now, the company believes GPT-5 will reestablish its dominance. "This is the best model in the world at coding," Altman stated. "It's also unmatched in writing and healthcare, among other areas."

Enhanced Capabilities and User Experience

The introduction of GPT-5 within ChatGPT presents it as a single model rather than separate ones for regular and reasoning tasks. A router developed by OpenAI automatically switches to a reasoning version for complex queries or when prompted to "think hard." Altman referred to the previous interface as a "very confusing mess."

Nick Turley, head of ChatGPT, expressed optimism about the new model's reception: "The vibes of this model are really good," he said. He believes that average users will particularly appreciate these improvements.

Availability and Customisation Options

GPT-5 is immediately available to all ChatGPT users. However, there is an undisclosed limit on prompts for free users; beyond this cap, a less powerful "mini" version is used. Developers accessing GPT-5 via OpenAI's API can choose from three versions: GPT-5, GPT-5 mini, and GPT-5 nano.

OpenAI has introduced four personality themes—"Cynic," "Robot," "Listener," and "Nerd"—to customise responses in ChatGPT. Users can also change chat thread colours for personalisation.

Advancements in Coding and Safety

Altman anticipates that GPT-5's coding prowess will lead to what he calls "software on demand." In tests, it outperformed other models on benchmarks like SWE-Bench and Aider Polyglot. During a press briefing, Yann Dubois demonstrated how GPT-5 generated a study website for learning French within seconds.

The safety research lead at OpenAI, Alex Beutel, noted that GPT-5 underwent over five thousand hours of testing to assess safety risks. A major focus was ensuring the model doesn't mislead users. Although it answers with fewer hallucinations than previous models, confidently lying remains an issue.

Challenges and Future Prospects

Beutel explained that safe completions allow GPT-5 to provide helpful answers while maintaining safety constraints. For example, if asked about energy needed to ignite materials—a potentially harmful query—it offers high-level information instead of detailed instructions.

OpenAI aims for GPT-5 to admit when it can't complete tasks or answer questions accurately. This transparency is intended to build user trust. The company hasn't disclosed specific data used for training GPT-5.

Altman acknowledges that while GPT-5 marks progress toward Artificial General Intelligence (AGI), it's still lacking continuous learning capabilities post-deployment. He remarked that this feature feels essential for AGI development.

"This is clearly a model that is generally intelligent," Altman stated regarding OpenAI's mission towards AGI development. Despite advancements with GPT-5, he noted it's still missing crucial elements needed for continuous learning from new information encountered during deployment.