OpenAI has unveiled GPT-5, its most advanced AI yet, which CEO Sam Altman calls “like talking to an expert in any topic.” The model is designed to handle complex reasoning with PhD-level proficiency, blending general Q&A, coding, and tool use in one system. Altman likened the jump from GPT-4 to GPT-5 to Apple’s shift from standard pixels to Retina Display.
Yet launch day brought an ironic hiccup: when solving 5.9 = x + 5.11, GPT-5 initially answered x = −0.21 instead of the correct 0.79. Engineers described this as a “reasoning slip,” where the AI mentally reversed subtraction order rather than calculating step-by-step. GPT models don’t “see” numbers as humans do; unless prompted for explicit calculations, they generate answers based on learned patterns, sometimes making sign or order mistakes.
Despite the error, GPT-5 shows major gains: OpenAI says it’s 45% less likely to produce factual mistakes than GPT-4o in standard mode and 80% less likely in its deeper “thinking mode.” It’s also more accurate in coding, capable of creating full websites, apps, or games from a single prompt, and better at following instructions precisely.
GPT-5 comes in three tiers — Mini, Regular, and Pro ($200/month) — with all signed-in ChatGPT users defaulting to it. The decimal slip, however, underscores that even top AI benefits from step-by-step prompting. As with any human expert, clear framing often separates a quick answer from a correct one.