OpenAI quietly dropped GPT-5 last night and the internet hasn't slept since. The new model isn't just an incremental upgrade — it represents a step-change in what AI assistants can reliably accomplish.
What's Different This Time
Unlike previous releases, GPT-5 ships with native multimodal reasoning baked in from day one. You can hand it a screenshot of your spreadsheet, describe what's wrong, and it'll write working Python to fix it without you touching a line of code.
The Business Case
For founders, the practical upshot is significant cost reduction in early-stage product work. Tasks that used to require a specialist — data analysis, copywriting, basic code debugging — can now be delegated to a well-prompted model.
What Stays Hard
The model still hallucinates. Context windows are larger but not unlimited. And the "reasoning" chains, while impressive, remain brittle when pushed into truly novel domains.