On August 7th, OpenAI released GPT-5. My feed immediately filled with benchmarks, pricing tables, and takes from people who had clearly been writing their posts before the model even shipped. So let me save you thirty minutes of scrolling: yes, it is better. Yes, it is cheaper. And no, that alone will not change your business.
GPT-5 arrived in three sizes, with a 400,000-token context window and pricing that undercuts its predecessor by 55 to 90 percent depending on the use case. The full model scores 74.9% on SWE-bench Verified, up from o3's 69.1%, while using 22% fewer output tokens. It shipped same-day into GitHub Copilot and Cursor. Within a week, reasoning use cases on the API jumped eightfold.
Those are real numbers. I don't want to downplay them. The model is genuinely better at multi-step reasoning, which means it can hold a longer chain of thought together, handle more complex instructions, and recover from mistakes mid-task.
The Wrong Gap to Watch
The thing I keep hearing is some version of "once the AI gets good enough, we'll start using it." GPT-5 closes the gap between AI that answers questions and AI that does work. But I've watched enough companies adopt AI to know that model capability was rarely the actual bottleneck.
The bottleneck is the workflow.
I talked to a brokerage last month that had been "waiting for the right model" to automate their proposal process. They had no documented workflow. No standard template. Three people doing the same task differently. A better model would have given them three different flavours of chaos, faster.
A better engine in the same broken car still doesn't get you anywhere useful.
What GPT-5 Actually Changes for a Small Company
If your workflows are already clean, this model lets you do things that weren't practical six months ago. Multi-step research tasks. Draft-review-revise cycles with fewer human checkpoints. Agentic coding where the AI writes, tests, and fixes its own code. Long-document analysis without chunking workarounds.
The pricing matters too. At $1.25 per million input tokens for the flagship model, cost is no longer a reasonable objection. You can run a customer support summarisation pipeline for pennies.
But none of that matters if you haven't done the boring work first. Map the process. Decide what "good" looks like. Build the feedback loop. Then plug in the model. The order matters.
What I Would Actually Do This Week
If you've been watching AI from the sidelines, GPT-5 is not the reason to start. The reason is that the cost of waiting compounds. Your competitors who started six months ago have working automations already.
Pick one workflow. Not the most complex one. The most repetitive one. Document what happens today. Then ask: where in this chain would a model that can reason through five or six steps actually help?
The model is genuinely impressive. I've been running it in my own coding workflows and the difference in multi-file reasoning is noticeable. But I had the workflows already. The model made good processes better. It did not rescue bad ones.
That's the one thing that matters. Better AI rewards better process design. It always has. It just rewards it faster now.