OpenAI’s latest ChatGPT model, GPT-5, has been caught making elementary mistakes in spelling and geography, despite its creators touting it as having “PhD level” intelligence. Users on social media quickly spotted the flaws after the model’s launch on Thursday.
One user on Bluesky found that GPT-5 repeatedly insisted there were three Bs in the word “blueberry”. The chatbot responded: “Yep – blueberry is one of those words where the middle almost trips you up, like it’s saying ‘b-b-better pay attention.’” Another user reported that the model could not correctly identify US states containing the letter R, and when asked to produce a map, it misspelled states including “Krizona” and “Vermoni”. It also invented states such as “New Jefst” and “Mitroinia”.
When Guardian Australia asked GPT-5 to identify the number of Rs in Australia’s states and territories, it correctly identified those that contain the letter, but mistakenly believed “Northern Territory” had only three Rs instead of five. When asked to produce a map, it spelled the territory as “Northan Territor”.
OpenAI CEO Sam Altman had described the new model as like having “access to a PhD-level expert in your pocket”, comparing it to previous versions that he likened to college and high school students. The company said on launching the product that there would be fewer errors and AI hallucinations under the latest model.
The source of the issues may lie in GPT-5’s architecture, which uses a “real-time router” to decide which of its AI models to use depending on the conversation type and intent. OpenAI has said that if you ask ChatGPT to “think hard about this”, it uses the latest reasoning model. Dan Shipper, chief executive of media and AI startup Every, found that the model would sometimes hallucinate on questions that should have been routed to the reasoning model. He noted that asking it to “think longer” often yielded accurate answers.
Despite the errors, GPT-5 was made available immediately to OpenAI’s 700 million weekly users. Altman said the AI had not yet reached artificial general intelligence, but described it as “generally intelligent” and a “significant step on the path to AGI”.
