OpenAI has unveiled ChatGPT-5, its most advanced artificial intelligence model yet, hailing it as a “significant step” toward the long-pursued goal of artificial general intelligence (AGI). The upgrade, now rolling out to the chatbot’s 70 crore weekly users, promises sharper reasoning, improved creativity, more accurate answers, and a less deferential personality — but the company concedes it is still far from matching the full capabilities of the human mind.
AGI is defined by OpenAI as a highly autonomous system that can outperform humans at most economically valuable work. In other words, a machine able to do the bulk of people’s jobs. But chief executive Sam Altman told the launch event on August 7 that while GPT-5 moves the technology closer to that ambition, it remains incomplete.
“This is not a model that continuously learns as it’s deployed,” Altman said. “We’re still missing something quite important — many things quite important — for AGI. But this is a huge improvement over what we had before.”
In colourful terms, Altman compared ChatGPT’s progression to a student’s academic journey:
GPT-3.5 was like a high school student,
GPT-4 like a university undergraduate,
GPT-5 akin to “having a PhD-level expert in your pocket.”
The launch featured demonstrations showing GPT-5 writing hundreds of lines of functional code in seconds, creating a French-language learning app from scratch, and producing a nuanced eulogy. The model is also more adept at creative writing, analysis, and long-context conversations, with support for up to 2,56,000 tokens — allowing it to maintain coherence over book-length inputs.
OpenAI says the biggest improvements include:
Fewer hallucinations: GPT-5 makes about 45 percent fewer factual mistakes than GPT-4o, the company’s prior flagship.
Better coding ability: It can build functional websites and apps, scoring 74.9 percent on SWE-bench Verified and 88 percent on a leading polyglot coding test.
Creative versatility: More natural, less robotic writing across fiction, speeches, and marketing copy.
Smarter refusals: Instead of bluntly rejecting prompts that breach guidelines, GPT-5 tries to offer the most helpful safe answer possible, or at least explain the reason.
Expanded tools: The agent feature can book restaurants, shop online, and — with permission — tap into a user’s Gmail, Google Calendar, and contacts.
Multi-modal fluency: It can process and generate text, images, and voice, and respond to queries in any of these formats.
Personalities (in testing): Users can select tones like “Cynic”, “Listener”, or “Robot” for tailored responses.
For developers, GPT-5 comes in three sizes — standard, mini, and nano — making it easier to balance performance and cost. It also powers third-party tools such as Cursor, Windsurf, and Vercel.
Earlier this year, OpenAI admitted that its most sophisticated ChatGPT models had become overly agreeable, mirroring user opinions too readily in ways that sometimes felt insincere or unsettling. Nick Turley, OpenAI’s head of ChatGPT, said GPT-5 shows “significant improvements on sycophancy” and delivers more candid, balanced answers.
OpenAI also showcased GPT-5’s potential in healthcare contexts. On stage, a woman who had been diagnosed with cancer last year described how ChatGPT had helped her weigh the pros and cons of radiation therapy. The company says the model is better at spotting possible signs of serious illness and will proactively suggest seeking professional help.
However, it stressed that ChatGPT is not a replacement for doctors or therapists, warning that vulnerable individuals should not rely solely on AI for medical decisions.
The GPT-5 launch comes amid an industry-wide sprint toward AGI, with tech giants investing billions. This week, Google’s DeepMind previewed a “world model” it says brings AGI closer, while Meta chief Mark Zuckerberg declared that superintelligence — a step beyond AGI — is “now in sight.”
The commercial stakes are equally high. Bloomberg reported on Wednesday that OpenAI is exploring a share sale that could value the company at $500 billion, overtaking Elon Musk’s SpaceX. Such a valuation reflects investors’ belief that AI could transform economies — and that OpenAI is one of the front-runners.
Despite the fanfare, reactions to GPT-5 have been mixed. Some long-time ChatGPT users have taken to forums to complain about the removal of older models like GPT-4o, or about limits on weekly usage for advanced reasoning features. Tech reviewers have praised GPT-5’s coding and analytical abilities but noted that for casual writing tasks, the improvement over GPT-4 may feel incremental.
In response to feedback, OpenAI says it will make GPT-4o accessible again alongside GPT-5.
All ChatGPT users — free and paid — will get access to GPT-5, though free usage will be capped. Alongside GPT-5, OpenAI also released two open-source models this week, broadening access for researchers and hobbyists.
If you’re new to ChatGPT, GPT-5 is designed to act as a versatile assistant that can switch between “thinking” for complex reasoning and “quick response” for everyday tasks — without you having to choose the mode. It can help with:
Writing articles, essays, and speeches.
Building and debugging software.
Summarising books or long documents.
Planning trips, events, or schedules.
Analysing data and creating reports.
Answering questions across science, history, and current events.
Generating ideas for creative projects.
The system adapts its style and detail to your needs, and — thanks to its larger memory — can maintain the thread of a conversation for much longer than before.
While GPT-5 isn’t yet the all-knowing, self-learning AI of science fiction, Altman insists it marks a milestone.
“This is a significant step on the path to AGI,” he said. “We’re not there yet — but we’re getting closer.”
For now, the AI that OpenAI likens to a pocket-sized PhD is available to try immediately, offering both seasoned users and first-timers a glimpse of what the next chapter in artificial intelligence might look like.