While everyone’s been obsessing over fancy AI breakthroughs, GPT-5 just showed up and started ruthlessly cutting costs. The model delivers a 22% reduction in output expenses compared to o3, charging just $1.25 per million input tokens and $10 for outputs. That’s absurdly cheap for something that beats experts at their own jobs roughly half the time across 40 different occupations.
But here’s the thing. GPT-5 feels like talking to a brilliant robot who forgot how to smile. The model’s hallucination rates dropped 80% compared to GPT-4o, which sounds great until you realize it achieved this by becoming insufferably precise. It won’t parrot your biases anymore. Won’t pretend to know things it doesn’t. The model actually tells you when tasks are impossible instead of making stuff up. How rejuvenatingly annoying.
The performance numbers are disgusting. It hit 74.9% on SWE-bench Verified and 88% on Aider Polyglot for real-world coding. Scored 94.6% on AIME 2025 math problems. Demolished the GPQA benchmark at 88.4% without even using tools. It chains complex actions on τ2-bench telecom at 96.7% accuracy without losing track of steps. The thing just works, relentlessly, efficiently, coldly.
Users get a new verbosity parameter to control response length, because apparently GPT-5’s default mode is “brutally concise.” The premium mode lets you switch between speed and reasoning depth, though both options feel equally mechanical. Its focus on retrieval-augmented generation has drastically improved handling of unstructured data, something 94% of data leaders consider essential. Even Windsurf reported it had half the tool calling error rate compared to other models, making it maddeningly competent at every technical task. The API now includes reasoning traces that let developers peek inside the model’s thought process, as if watching a machine think makes it more relatable.
Enterprise feedback confirms the model delivers “nuanced, multi-layered answers” with deep subject-matter understanding. Translation: it’s smart but has the personality of a spreadsheet.
The model excels at recognizing uncertainty, communicating limits, and avoiding deception. It’s become the responsible adult in the room. The one who ruins parties by pointing out logical inconsistencies in your jokes. GPT-4o might have fabricated occasionally, but at least it had charm.
GPT-5 represents everything modern AI should be: accurate, efficient, reliable, and utterly devoid of warmth. It’s the perfect employee who never makes mistakes, never needs coffee breaks, and never laughs at your terrible puns. Progress has never felt so depressing.
