OpenAI released GPT-5 on August 7, 2025, and the practical change at your desk is subtraction: the model picker largely disappears, with the service routing requests automatically between fast answers and slower deep reasoning, per OpenAI's launch announcement and Reuters' August 2025 coverage. Free users got GPT-5 access with rate limits from day one, while paying Plus users received higher caps. RechargeMe covers this as sourced news, not hands-on testing — we haven't run a prompt on it, and vendor claims are labeled as such.
Per OpenAI's own materials, the company reported improved performance over GPT-4o on its internal benchmarks, including its SWE-bench Verified coding evaluation. Vendor benchmarks are exactly that: chosen by the vendor, on the vendor's test set. Independent verification at launch was thin, and that remains the honest state of the evidence.
What's the overlooked detail?
The thinking-time dial. GPT-5 lets users set how long the model reasons — and on paid tiers, more thinking consumes a larger share of your weekly usage allowance, per OpenAI's post-launch help documentation from August 2025. The quiet consequence: your subscription's effective capacity now depends on how you set a slider, not just how many messages you send. Two users on the same plan can burn vastly different quota with identical prompt counts. That change showed up in help pages, not the keynote, and it's the part most likely to surprise people mid-week.
What the announcement did not establish: real-world reliability gains in everyday drafting and summarizing work. The benchmarks cover code and reasoning puzzles, not your Tuesday inbox. Independent test results will tell that story over the coming months; until then, the documented facts are the routing, the rationing, and the slider — and the pricing page, as always, is the part worth reading twice.