GPT-5 Arrives — Brilliant, and a Little Bumpy

13 August 2025

OpenAI has launched GPT-5, its long-awaited flagship, now powering ChatGPT as a single unified system that decides for itself when to answer quickly and when to think harder. The gains in coding, reasoning and technical work are genuine and substantial.

The launch itself? Less smooth. Users complained loudly about model changes and inconsistencies, and Sam Altman publicly admitted the rollout had been mishandled in places. Even at the frontier, shipping is hard.

Three lessons for the rest of us

1. Don't build on sand you haven't tested. If your business processes depend on an AI tool, version changes can alter your results overnight. Keep examples of "known good" outputs so you can spot drift.

2. The capability curve hasn't flattened. Each generation makes previously impractical tasks routine. Whatever you concluded about AI's limits a year ago is due for review.

3. Cheaper tiers are coming everywhere. Alongside GPT-5, OpenAI introduced a budget subscription tier — a sign the industry is racing to make capable AI near-universal. Your competitors' access to these tools is no longer a budget question. Only the skill gap remains.

The takeaway

Judge AI models like employees: not by their interview (the launch demo) but by their probation period. Give GPT-5 your real tasks for a fortnight and let the results, not the headlines, decide.