OpenAI’s GPT-5 debuted with strong benchmark claims, but early reviews were sharply divided. Tests praised its coding and logical reasoning while criticizing its creative writing, information retrieval, and even basic arithmetic reliability.

OpenAI released GPT-5, claiming it's the 'smartest, fastest model ever' with benchmark scores of 94.6% on math and 74.9% on real-world coding. However, early user reactions were sharply divided: Reddit threads calling it 'horrible' and 'underwhelming' gained thousands of upvotes, and over 3,000 people signed a petition to bring back GPT-4o. Polymarket odds for OpenAI having the best model by end of August collapsed from 75% to 12%. Our comprehensive tests reveal that GPT-5 excels in coding and logical reasoning—at a fraction of the API cost of Claude 4.1 Opus—but fails spectacularly at elementary arithmetic, creative writing, and long-context information retrieval. We compare it directly with Claude, Grok, and Gemini, dissecting its strengths and weaknesses. The verdict: GPT-5 is a powerhouse for developers and analysts, but a step back for casual users and storytellers.