Incident: The router broke and answers got worse

Module 13 · Incident file · About 30 minutes

What happened

On 7 August 2025 OpenAI launched GPT-5 in ChatGPT as one system with several models behind it. Its system card describes "a smart and fast model that answers most questions, a deeper reasoning model for harder problems, and a real-time router that quickly decides which model to use based on conversation type, complexity, tool needs, and explicit intent" (OpenAI, GPT-5 system card, 7 August 2025).

The next day Sam Altman posted an update on X, quoted in full by Simon Willison (Willison, 8 August 2025): "Yesterday, the autoswitcher broke and was out of commission for a chunk of the day, and the result was GPT-5 seemed way dumber." He promised "interventions to how the decision boundary works", to "make it more transparent about which model is answering a given query", and to make it "easier to manually trigger thinking". The same post said API traffic had "about doubled over the past 24 hours".

BleepingComputer's Mayank Parmar reported user complaints the same day and that GPT-5 "often fails to switch to its new reasoning capabilities when users expect" (BleepingComputer, 8 August 2025). OpenAI published no postmortem: no duration in hours, no count of users affected, no cost.