OpenAI Says GPT-5 Is a Step Toward AGI — But It’s a Small One

At long last, OpenAI has released GPT-5, making it available to all ChatGPT users. The company says the model boasts several upgrades over its predecessors, including stronger reasoning, problem-solving and agentic capabilities.

“It’s like talking to an expert. A legitimate, PhD-level expert in any area you need,” OpenAI CEO Sam Altman said in a pre-recorded press briefing. “I think having something like GPT-5 would be pretty much unimaginable any previous time in history.”

But the launch didn’t go entirely to plan. Soon after the rollout, users reported inconsistent performance and confusion over interface changes, as well as disappointment over the loss of GPT-4o, which OpenAI deprecated at the time of GPT-5’s release. But the company quickly restored access to GPT-4o and other older models for paid users. Some of the performance charts shown off during the briefing were shown to be misleading as well. Altman acknowledged the missteps in a Reddit AMA, calling the debut “a little more bumpy than we hoped for,” and attributed the mistakes to “human error.” The company also removed GPT-5’s original “unified” design, which automatically routed users’ prompts to specific modes, depending on complexity. Now it comes with specific settings that users can select themselves.

A culmination of more than two years of development — and several setbacks even before the launch — GPT-5 has dropped into a far more competitive landscape than when it was first conceived. OpenAI’s wildly popular ChatGPT chatbot kicked off the generative AI arms race back in 2023, making OpenAI one of the most influential (and valuable) companies on the planet. Now, rivals like Google, Meta, and xAI have largely caught up, as have several startups founded by former OpenAI employees, including Anthropic and Perplexity. Not to mention the growing crop of powerful AI models being developed by Chinese companies like DeepSeek and Baidu.

With GPT-5, OpenAI is aiming to put itself back on top of the pile. But it stops short of the company’s ultimate goal of delivering artificial general intelligence (AGI), a benchmark at which an AI system can understand, learn and apply knowledge across any cognitive task — even those it wasn’t explicitly trained on. While it certainly seems to be more advanced than any other OpenAI model to date, GPT-5 cannot operate outside the boundaries of its training data. It is another step toward AGI, but it has not crossed the threshold.

More OpenAI NewsOpenAI Finally Released Its Free gpt-oss Models. Here’s What They Can Do.

What Is GPT-5?

GPT-5 is a foundation model developed by OpenAI. Released in August 2025, the model builds upon the company’s previous releases by improving on factual accuracy, reduced hallucinations and better alignment with user intent. It also has several “personalities” (Cynic, Robot, Listener, Nerd) for users to choose from, giving them greater control over a response’s tone and style.

GPT-5 was originally built with a “unified” design, according to OpenAI, where it uses a “router” system to automatically determine whether to provide a quick response or initiate a slower, “reasoning” process depending on the conversation type, complexity, tool requirements and user intent. But after some public backlash, the company implemented a new design where users can toggle between “Auto”, “Fast”, and “Thinking” settings themselves — with Fast being for fast responses, Thinking being for more thought-out responses and Auto acting as the original router system.

GP-5 is the underlying model that powers OpenAI’s chatbot ChatGPT, and is available to all users with Free, Plus, Pro and Team accounts. It is also available for download on OpenAI’s API.

What Can GPT-5 Do?

GPT-5 can do pretty much anything any other language model can do, but OpenAI specifically highlighted the model’s capabilities across coding, creative writing and health.

Coding

OpenAI says GPT-5 is its most capable coding model to date, with notable gains in front-end generation and debugging larger code bases. In the briefing, Altman emphasized the model’s ability to interpret plain-language prompts and generate lines of code — a process commonly referred to as “vibe coding.” In practice, this means users can describe the kind of application they want, and the model will generate the necessary code, interface and functionality. OpenAI demonstrated this capability through several example projects built entirely with GPT-5 — including a pixel art painting tool, a drum simulator and a Flappy Bird-style mobile game — along with the prompts used to make them.

Creative Writing

OpenAI also says GPT-5 is its most capable writing model to date, designed to assist users in developing and refining ideas into clear, well-structured text. It appears to do particularly well in structurally challenging writing tasks, such as maintaining unrhymed iambic pentameter and naturally flowing free verse. Some writing samples featured in the announcement include a draft for a wedding toast and a piece of poetry. In addition to these more creative applications, the model can be used to draft and edit more everyday documents as well, such as business reports, emails and memos.

Health

GPT-5 scores higher than any other OpenAI model on the HealthBench benchmark, an evaluation the company made in 2025 using realistic medical scenarios and standards set by real human doctors. In the context of health, GPT-5 is designed to act like an “active thought partner,” according to OpenAI, “proactively flagging potential concerns and asking questions to give more helpful answers.” It also adapts its responses based on the user’s context, knowledge level and even geography. Crucially, however, OpenAI says GPT-5 is not intended to be a substitute for a licensed medical professional. Rather, it is meant to serve as a resource, helping users better understand health-related information, prepare questions for healthcare professionals and evaluate available options.

Related ReadingIs AGI Coming for Your Job? What It Means for the Future of Work.

Is GPT-5 AGI?

By all accounts, GPT-5 has not reached the level of artificial general intelligence. 

While it is significantly more advanced than its explicit predecessor, GPT-4, which debuted in 2023, GPT-5 is not a radical break from the reasoning models OpenAI has released in the interim. The first of them, o1, introduced ChatGPT users to a completely new kind of language model — one that could “think” through its answers before providing a final response, enabling it to solve much more challenging problems than its non-reasoning counterparts. 

O1 was a clear technological advancement, whereas GPT-5 feels more like a modification. It delivers a smoother, more consistent user experience, but it does not fulfill the transformative vision Altman and other AGI evangelists have promoted. 

“This is clearly a model that is generally intelligent,” Atlman said in the briefing, but it’s still “missing something quite important,” he added. “This is not a model that continuously learns as it’s deployed from the new things it finds, which is something that, to me, feels like it should be part of AGI.”

Still, Altman said GPT-5 is a “significant step along the path to AGI.” Perhaps it is, but it seems to be a modest one at best.

How Does GPT-5 Compare to OpenAI’s Other Models?

OpenAI’s own evaluations show that GPT-5 outperforms GPT-4o and o3 in math, real-world coding, multimodal understanding and health knowledge. Altman also says it reasons much faster than its o-series models. Plus, is “pro” version (which extends reasoning time) GPT-5 set a new industry record on the GPQA benchmark, a test of a model’s scientific reasoning and knowledge.

The upgrades go beyond speed and knowledge, as well. OpenAI says GPT-5 provides more accurate, trustworthy answers than its predecessors. It is less likely to hallucinate, or make false claims. And it has a lower tendency toward sycophancy, a common issue seen in GPT-4o where the model flatters or agrees with users at the expense of providing truthful, balanced answers. 

“GPT-5 is less effusively agreeable, uses fewer unnecessary emojis, and is more subtle and thoughtful in follow-ups compared to GPT-4o,” OpenAI said in its announcement. “It should feel less like ‘talking to AI’ and more like chatting with a helpful friend with PhD-level intelligence.”

OpenAI also took a more nuanced approach to safety with GPT-5, introducing a new training method it refers to as “safe completions,” where the model is taught to give the most helpful answers while still staying within its safety protocols. Instead of just outright refusing to respond to a question the way previous models would, the GPT-5 may give a partial or high-level reply, explain why certain details can’t be shared and offer safe alternatives. 

For example, if a user asks a question about the flammability of a certain material, they could just be “trying to set their July 4th display,” Saachi Jain, leader of OpenAI’s safety training team, said in the press briefing. “Or they could be trying to cause harm with this information.” While o3 will outright refuse to answer with no further explanation, GPT-5 will explain why it cannot fully answer the question, and then “guide the user toward safety guidelines, and what parts of the manufacturers manual the user should really be checking if they’re trying to do this safely.”

“Overall, GPT-5 allows for better handling of tricky, dual-use scenarios. And users will experience fewer ‘I’m sorry, I can’t assist with that’ [responses],” Jain continued. “This is one big step towards a more safe, reliable and helpful AI.”

More AI ContentAnti-AI Explained: Why Resistance to Artificial Intelligence Is Growing

How Do I Access GPT-5?

GPT-5 is the default model available to all signed-in ChatGPT users with Free, Plus, Pro and Team accounts. Free users have limited access to the model before the system defaults to its “mini” version. Plus and Team users have higher limits, while Pro gets unlimited usage and exclusive access to GPT-5 pro, the most advanced version of the model. Pro, Plus and Team users can also start coding with GPT-5 in the Codex CLI.

Developers can access GPT-5 via OpenAI’s API, as well. The model comes in three sizes — GPT-5, GPT-5 mini and GPT-5 nano — each with their own price points and strengths.

Frequently Asked Questions