OpenAI CEO Sam Altman addressed user concerns about the GPT-5 rollout during a Reddit AMA on August 8, 2025.

Some users said GPT-5 performed worse than GPT-4o due to issues with a new real-time router meant to select the best model for each prompt.

Altman explained a technical problem with the router made GPT-5 appear less capable at launch but said fixes had been made and further improvements are planned.

He also said OpenAI is considering allowing Plus subscribers to keep using GPT-4o after strong user demand.

Rate limits for Plus users would double as the GPT-5 rollout continues.

The AMA included questions about an inaccurate chart shown in GPT-5’s launch presentation, which Altman called a “mega chart screwup” on X.

Altman said OpenAI would keep working to improve GPT-5 and respond to user feedback.

🔗 Source: TechCrunch

🧠 Food for thought

1️⃣ Technical complexity behind AI routing systems creates rollout vulnerabilities

GPT-5’s router problems highlight how sophisticated AI systems introduce new failure points that can dramatically affect user experience.

The model features a “real-time router” that automatically switches between fast and deep thinking modes based on user needs 1. However, when this system failed during rollout, users perceived the AI as significantly less capable than its predecessor.

This technical architecture represents a shift from single-model systems to complex decision-making frameworks that must work seamlessly to maintain user trust.

The router’s failure created a disconnect between GPT-5’s benchmark performance—achieving 89.4% accuracy on Graduate-Level Science Questions and 100% on mathematical reasoning tests 1, 2—and real-world user experience during the outage.

This highlights how advanced AI systems can fail not due to their core capabilities, but because of the sophisticated orchestration layers designed to optimize their performance.

2️⃣ User feedback drives immediate product decisions in competitive AI markets

Recent OpenAI developments