GPT-5 wants to be your coder, tutor and thought partner

OpenAI has officially launched GPT-5, describing it as “our smartest, fastest and most useful model yet, with thinking built in.” The new flagship model is now available to everyone through ChatGPT and across OpenAI’s API platforms.

According to OpenAI, GPT-5 provides “expert intelligence for everyone.” The model, the company says, is “smarter across the board” and aims to deliver “more useful responses across math, science, finance, law and more.” OpenAI compares it to “having a team of experts on call for whatever you want to know.”

The release is part of a broader summer update from OpenAI and reflects continued development toward more contextually aware, versatile and personalized AI systems. Users can “try it in ChatGPT,” while developers and businesses are invited to “start building” with GPT-5 through the OpenAI API.

GPT-5 also comes with new features in ChatGPT. One such feature is customization. Users can select a personality from four preset personality types that might better align with their communication style: Cynic, Robot, Listener and Nerd. Meanwhile, a new voice feature “understands your instructions better and allows you to tailor its speaking style.”

A “study mode” provides “personalized, step-by-step help to learn anything,” while integration options allow users to “connect your Gmail and Google Calendar” so the model can generate personalized responses while “respecting existing permissions.”

Finally, OpenAI describes GPT-5 as its most advanced model to date for coding and agentic tasks, with capabilities including high-quality code generation, front-end UI creation with minimal prompts and enhanced performance in personality modeling, steerability and executing extended sequences of tool calls.

Shobhit Varshney, Head of Data and AI at IBM, praised the release, describing it in a LinkedIn post, as “the best model for coding and agentic tasks across domains.” He noted that the system introduces multiple levels of reasoning effort—high, medium, low and minimal—allowing it to scale its “thinking” based on task complexity and cost.

Varshney also highlighted the model’s real-world training. “They have had millions of real-world interactions (both naive and rogue) to train on,” he wrote. “This is where a proprietary hosted SaaS model has a competitive advantage.”

While the flagship GPT-5 model may have generated the most buzz, it was the distilled smaller versions, particularly GPT-5 nano, that impressed IBM Distinguished Engineer Chris Hay. “This outperforms most of the large models in the market, especially with agentic tasks and tool calling,” he said on a recent episode of Mixture of Experts.

The new model router also makes it dramatically easier for people to select the GPT model best suited to their task. “One of the memes on the internet has been how complicated it has been to go into [OpenAI’s] model selector and figure out what model you're supposed to use for anything,” said Hay. “Consolidating everything in the one brand family around GPT-5 with a model router in front of it was their way of delivering against that.”

The price and reliability of GPT-5 was most noteworthy to Mihai Criveti, a Distinguished Engineer of Agentic AI at IBM. “It’s particularly good at tool calling, which is critical for agents,” he said on the Mixture of Experts episode. “It’s much more reliable and sells at a cost that is sustainable for these types of agentic workloads.”

On the topic of coding improvements—one of the chief innovations being touted by OpenAI with the GPT-5 release—Hay and Criveti were more muted. Criveti got access to GPT-5 around 2 a.m. on Friday, August 8. “I got out of bed. I rushed to my machine, and I was working all night, trying out prompts, trying it with different tools, trying it with MCP [Model Control Protocol],” he said. “And in the morning, I had to start my day job, and I found myself using Claude Opus 4.1 again,” he continued, referring to Anthropic’s flagship coding model. Hay was in the same boat in terms of whether GPT-5 would replace his tried-and-true LLMs for day-to-day coding tasks. “I had high hopes, but no.”

PJ Hagerty, Lead AI Advocate for IBM, praised the new release as a leap forward for day-to-day business operations. “The biggest value add for enterprises is contextualizing the massive amount of content they have to deal with and making them succinct in a way that people can index them,” he said in an interview with IBM Think.

He was skeptical, however, about some of the claims the OpenAI team made during Thursday’s big announcement, including founder Sam Altman’s assertion that GPT-5 talks like a “legitimate PhD-level expert.”

“It does not,” said Hagerty. “It reads academic theses. I could read theses, but I don’t have the lived experience of someone with a PhD, or the practical skills. If I read all the Twilight books, it doesn’t mean I know more than anyone else about vampires.” 

Despite these reservations, Hagerty is bullish on what ChatGPT signals for the future. He said, “It’s bringing us closer to general AI”—the as-yet-theoretical stage at which AI can replicate human intelligence.

Hagerty also sees OpenAI’s latest offering as being motivating for the large language model (LLM) industry as a whole. “This will put pressure on a lot of other LLMs to up their game,” he said. “In the way that when the first iPhone came out and Android had to compete, that’s what we’re seeing now.”