TL;DR
OpenAI disabled automatic routing of free users to expensive reasoning models, reverting to faster GPT-5.2 Instant as default after user engagement declined.
Key Points
- Model router increased free user reasoning model usage from <1% to 7% in four months before rollback
- Feature negatively impacted daily active users metric; users preferred speed over answer quality
- Paid Plus ($20/mo) and Pro ($200/mo) tiers retain model router; free/Go users must manually select reasoning models
- GPT-5.2 Instant now enhanced with increased safety performance and longer reasoning capabilities to narrow performance gap
Why It Matters
This reveals the fundamental tension in consumer AI products: inference latency and cost often trump reasoning quality for mass-market adoption. For infrastructure engineers and ML ops teams, it underscores the challenge of cost-effectively scaling reasoning models at consumer scale, and suggests that automatic model selection at the router level remains unsolved for free-tier users.
Source: www.wired.com