TLDR: OpenAI CEO Sam Altman has publicly acknowledged significant issues with the GPT-5 launch, including technical glitches, a ‘robotic’ user experience, and misleading performance metrics. User backlash led to the reinstatement of older models and a commitment to improve future AI iterations, shifting focus towards GPT-6.
OpenAI’s highly anticipated GPT-5 model, launched in mid-2025, has faced a tumultuous rollout, prompting an unusual public admission of shortcomings from CEO Sam Altman. The company has reportedly disclosed the underlying reasons for the model’s perceived limitations, which have led to widespread user dissatisfaction and a preference for its predecessor, GPT-4o.
The launch was marred by a series of technical and experiential issues. A significant point of contention was a misleading performance chart presented during the live-streamed launch on August 8, 2025. The chart, intended to showcase GPT-5’s accuracy, featured misaligned bars. Altman later clarified that the actual SWE-bench Verified data indicated GPT-5’s accuracy at 74.9% with a ‘thinking mode’ and 52.8% without, a notable improvement over GPT-4o’s 30.8%, but the initial presentation caused confusion and online ridicule.
Beyond data presentation, users reported a ‘robotic, rigid, and dehumanized’ tone in GPT-5’s responses, a stark contrast to the more fluid and empathetic interactions offered by GPT-4o. This perceived loss of user-friendliness was a major complaint, with many users expressing disappointment and frustration. Petitions circulated online, demanding the reinstatement of older models, leading OpenAI to quietly restore access to GPT-4o and other legacy models for ChatGPT Plus subscribers.
Technical glitches further compounded the problems. The ‘autoswitcher,’ a system designed to select the optimal model for a given request, proved problematic, often delivering incoherent answers and leading to unpredictable response times. Reports also highlighted bugs and broken features, including issues with Custom GPTs and the deep research mode. Perhaps most surprisingly, GPT-5 reportedly struggled with basic arithmetic, failing to solve simple second-grade math equations, raising questions about its ‘PhD-level expert’ claims.
OpenAI’s own data indicated that GPT-5 still made incorrect claims 9.6% of the time, though this figure reportedly dropped to 4.5% with the ‘thinking’ mode enabled. The rapid doubling of API traffic within 24 hours of launch also exposed scaling and performance limitations, resulting in power caps, increased token costs, and inference delays that impacted platform stability.
Adding to the concerns, security firm SPLX identified vulnerabilities to prompt injection and logic attacks within GPT-5. These issues, coupled with growing competition from rivals like Google and Anthropic, have intensified pressure on OpenAI to demonstrate GPT-5’s revolutionary capabilities.
Also Read:
- OpenAI Halts Disrespectful Deepfakes of Martin Luther King Jr. on Sora Platform
- AI Ethicist Timnit Gebru Raises Alarms Over Industry Practices, Citing Data Exploitation, Environmental Impact, and ‘Erotica’ Monetization
In response to the backlash, Sam Altman acknowledged that the company ‘totally screwed up some things on the rollout,’ admitting to mismanagement of communication and expectations. OpenAI has pledged to develop new features for future GPT iterations to address the ‘colder, less human tone’ and is already shifting its focus towards GPT-6, which Altman claims will be a ‘better model’ more aligned with user desires. The company’s ability to adapt to this feedback will be crucial in maintaining its leadership in the rapidly evolving AI landscape.
