KAMPALA – Last Thursday morning, OpenAI quietly crossed a threshold it’s been inching toward for years. GPT-5, the company’s most advanced artificial intelligence model yet, went live for 700 million weekly ChatGPT users. In an instant, the chatbot that once answered trivia and drafted polite emails was transformed into something closer to a digital assistant that can plan your week, code your app, and, if you let it, make decisions on your behalf.
Sam Altman, OpenAI’s CEO, didn’t mince words in the launch briefing. GPT-5, he said, is “the best model in the world,” a leap forward on the company’s path toward building artificial general intelligence, machines that can outperform humans at most economically valuable work. “Having something like GPT-5,” Altman said, “would be pretty much unimaginable at any previous time in history.”
This isn’t just marketing bravado. GPT-5 isn’t a single skill upgrade; it’s a structural overhaul, the first “unified” model from OpenAI. It merges the company’s two main AI lines: the fast, conversational GPT series and the slower but more deliberate “o-series” reasoning models. The result is an assistant that can decide for itself whether to give you an instant answer or pause to think more deeply before replying.
When ChatGPT debuted in late 2022, it was hailed as a breakthrough in human-like conversation. People used it to brainstorm, summarize news, and debug code. But the tool was reactive, you asked, it answered. GPT-5 pushes past that passivity.
Now, ChatGPT can autonomously generate fully functional software from a plain-language request, navigate your online calendar, prepare research briefs, or extract the relevant details from a 200-page contract without you needing to break it into chunks. The difference is subtle in a single exchange, but striking over time: GPT-5 doesn’t just respond, it works.
Nick Turley, OpenAI’s vice president for ChatGPT, called this the first time free users have had access to the company’s advanced reasoning models. “This is just one of the ways that I’m excited to live the mission,” he said, referencing OpenAI’s aim to distribute powerful AI widely, not just to paying customers.
The Benchmarks—and the Stakes
OpenAI claims GPT-5 is now the best coding model in the world. On SWE-bench Verified, a benchmark of real GitHub coding tasks, it scores 74.9 percent on its first attempt, narrowly beating Anthropic’s Claude Opus 4.1 at 74.5 percent, and far ahead of Google DeepMind’s Gemini 2.5 Pro at 59.6 percent, according to TechCrunch. In health information, one of the most sensitive domains for AI, GPT-5’s error rate is just 1.6 percent on difficult medical queries, compared to over 12 percent for GPT-4o.
In creative tasks, Turley says the model simply has “better taste.” It writes more fluidly, designs more elegantly, and in OpenAI’s internal evaluations, gets closer to what human evaluators actually like.
The competition is watching closely. Anthropic, Google, and Elon Musk’s xAI have all released their own frontier models this year. On some benchmarks, GPT-5 underperforms rivals: in “Humanity’s Last Exam,” which measures performance across math, science, and humanities, GPT-5 Pro scored 42 percent, slightly less than xAI’s Grok 4 Heavy at 44.4 percent. But in PhD-level science questions, it took the crown with an 89.4 percent score.
For OpenAI, these decimal-point leads and losses matter less than the bigger picture: GPT-5 is meant to feel like an agent, not a search box. On Tau-bench, which tests how well AI can navigate websites to complete tasks, it handled retail sites with over 81 percent accuracy. Airline booking was trickier; it scored 63.5 percent, just under the older o3 model. Still, the company says the ability to take multi-step actions on its own will improve as more people use it.
Making the Assistant Yours
Alongside the model itself, ChatGPT gets four new “personalities”: Cynic, Robot, Listener, and Nerd. These aren’t gimmicks so much as presets for tone and style, letting people nudge the assistant toward dry humour, clinical precision, quiet empathy, or boundless enthusiasm without re-prompting every time.
For developers, GPT-5 arrives in three sizes, standard, mini, and nano, each balancing speed against depth of reasoning. The Pro tier taps extra computing power to push accuracy and creativity further. API pricing starts at $1.25 per million input tokens, making it cheaper than training most custom models from scratch.
Why This Matters Beyond Tech
To most people, the upgrade will be invisible at first. Ask GPT-5 to draft an email, and it will, just as GPT-4o did. The difference is in what happens when you push further: “Book me a hotel in Nairobi that meets these five criteria and send the receipt to my finance app.” Or: “Write and deploy a Chrome extension that blocks addictive websites during my work hours.”
Tasks like these are the reason many in Silicon Valley see GPT-5 as a bellwether for AI’s trajectory. If GPT-4 proved that AI could talk like us, GPT-5 is a test of whether it can reliably act for us, in workplaces, in creative industries, and in daily life. With 700 million people already using ChatGPT each week, even small shifts in behaviour could ripple across economies.
Safety concerns persist. AI “hallucinations,” false or fabricated answers, have dropped from over 20 percent in GPT-4o to under 5 percent in GPT-5, but they haven’t disappeared. OpenAI says deception rates are also down, and the model is better at spotting malicious requests without blocking harmless ones. But the more an AI acts autonomously, the higher the stakes when it makes a bad call.
Alex Beutel, who leads safety research at OpenAI, framed it as a trust issue. “Reducing deception makes the model not just safer, but more transparent and honest in ways users can trust,” he said.
The Road Ahead
Two years ago, ChatGPT was a curiosity that students used for essays and programmers used for code snippets. Today, it’s edging toward something else entirely: a co-worker, a research assistant, or a digital fixer that might one day anticipate needs before you articulate them.
Altman’s vision of AI that can “outperform humans at most economically valuable work” still raises more questions than it answers. What happens when such a system is in everyone’s pocket? Who benefits most, individuals, corporations, or the model’s creators? And how do we keep it from quietly making choices we didn’t ask it to make?
Those questions won’t be resolved by this release. But GPT-5 makes one thing clear: the age of the chatbot is ending. What comes next will be more capable, more independent, and, for better or worse, more woven into the way we live and work.
