TLDRs:
- GPT-5 rollout faced technical glitches with its real-time router, leading users to report reduced performance.
- OpenAI plans to fix issues and may allow Plus subscribers to keep access to GPT-4o.
- Rate limits for Plus users will double during the GPT-5 rollout to improve usability.
- Altman vows more transparency and quicker responses to user concerns following strong feedback.
OpenAI CEO Sam Altman has confirmed that a technical problem affected the performance of GPT-5 during its rollout, prompting swift fixes and promises of greater user control.
Speaking during a Reddit AMA on Friday, Altman responded to concerns that GPT-5 was underperforming compared to GPT-4o.
The issue stemmed from GPT-5’s “real-time router,” a system designed to automatically choose between different reasoning modes, quick responses for simple prompts and deeper analysis for complex tasks. According to Altman, the router malfunctioned during launch, leading users to perceive GPT-5 as less capable despite its strong benchmark scores.
Behind the Technical Failure
Notably, the router’s failure didn’t mean GPT-5 lacked intelligence. In fact, the model achieved 89.4% accuracy on Graduate-Level Science Questions and perfect scores on mathematical reasoning tests.
However, because the orchestration layer faltered, these capabilities didn’t translate into real-world interactions.
This incident underscores the vulnerabilities in advanced AI architectures. While single-model systems rely on consistent outputs, GPT-5’s multi-model routing demands flawless coordination. When that coordination breaks, user trust can suffer, even if the underlying model remains powerful.
Listening to the Users
Altman emphasized that user feedback is shaping OpenAI’s immediate strategy. Many Plus subscribers expressed a preference for GPT-4o during the rollout, leading OpenAI to consider making it available alongside GPT-5. “We’ve heard you,” Altman said, noting that technical innovation must be balanced with familiarity and reliability.
In addition, OpenAI will double rate limits for Plus users during the transition. This move aims to enhance productivity for users who rely on the service for heavy workloads, while ongoing updates to GPT-5 address early performance issues.
The AMA also touched on a separate controversy: an inaccurate chart shown during GPT-5’s launch presentation. Altman referred to it as a “mega chart screwup” on X (formerly Twitter) and pledged more rigorous review processes for future product announcements.
Adapting in Real Time
OpenAI’s rapid response reflects the increasingly competitive nature of the AI market. With rivals like Anthropic, Google DeepMind, and Meta rolling out their own large language models, user satisfaction is crucial. Any perceived drop in performance risks shifting users to competing platforms.
Altman’s approach suggests OpenAI is treating model launches less like one-time events and more like evolving conversations with its community. By combining technical fixes, policy adjustments, and direct engagement, the company aims to turn a shaky rollout into a case study in responsive product management.
While the GPT-5 router glitch was a setback, Altman framed it as an opportunity to refine both the technology and the relationship with new users.
