OpenAI’s GPT-5 has landed to a split verdict: early testers have reported steady upgrades in coding and multi-step reasoning, but many users have complained about hallucinations and slip-ups that dulled the “seminal” launch moment CEO Sam Altman had primed.
Developer Simon Willison called GPT-5 “competent” and “occasionally impressive”, yet “not a dramatic departure” in an interview with Bloomberg News.
Part of the confusion stems from how GPT-5 works: it automatically routes queries across different model variants. Altman said a day-one glitch in that “auto-switcher” made GPT-5 seem worse than it is and that performance should already look stronger.
The stakes are huge. OpenAI is racing rivals in the US and China while trying to convert usage into revenue at massive scale; the company now counts roughly 700 million weekly ChatGPT users, underscoring why perceptions of GPT-5 matter.
Read: OpenAI drops first open models since GPT-2
On benchmarks, GPT-5 quickly climbed user-voted LMArena leaderboards, but on the ARC-AGI-2 test it lagged top competitors such as xAI’s latest Grok – another reason reactions were mixed pending deeper evaluation. – © 2025 NewsCentral Media
Get breaking news from TechCentral on WhatsApp. Sign up here.
Don’t miss:
OpenAI unveils GPT-5, promising ‘on-demand software’ and PhD-level expertise
