OpenAI's GPT-5 Launch Plagued by Performance Issues

The highly anticipated launch of OpenAI’s GPT-5 model has encountered a challenging reception, marked by a series of user-reported errors and widespread dissatisfaction. Following its livestreamed presentation, which itself included glitches in charts and voice demonstrations, users quickly began flagging instances where GPT-5 faltered on tasks that earlier OpenAI models, and even rival AI systems, handled correctly.

Notable examples of these missteps include GPT-5’s failure to correctly solve basic mathematical problems. Data scientist Colin Fraser shared screenshots illustrating the model’s inability to prove whether the repeating decimal 8.888 is equal to 9 (which it is not). It also stumbled on a straightforward algebraic equation, 5.9 = x + 5.11, a problem elementary school students could typically solve. Furthermore, attempts to use GPT-5 to evaluate OpenAI’s own erroneous presentation charts yielded unhelpful and incorrect responses. The model even struggled with a more complex math word problem that, while initially tricky for humans, was accurately solved by Elon Musk’s Groq 4 AI.

Beyond mathematical reasoning, GPT-5’s coding prowess, despite strong internal and some third-party benchmarks, appears to fall short in real-world application. While benchmarks suggested superiority, developers report that Anthropic’s recently updated Claude Opus 4.1 often performs better at “one-shotting” tasks—completing a user’s desired application or software build in a single attempt. This disparity highlights a gap between theoretical performance metrics and practical utility.

Compounding user frustration, OpenAI is gradually deprecating its older, well-regarded models, including GPT-4o and the powerful reasoning model o3, for ChatGPT users. While these will remain accessible via the application programming interface (API) for developers, their removal from the primary user interface has been met with dismay. Adding to the concerns, a report from security firm SPLX indicated that OpenAI’s internal safety layers exhibit significant vulnerabilities, particularly in areas like business alignment and susceptibility to prompt injection and obfuscated logic attacks.

Early feedback from the AI community paints a picture of a lukewarm reception. A poll conducted by AI influencer Bilawal Sidhu asking for a “vibe check” on GPT-5 saw an overwhelming majority describe it as “Kinda mid.” This sentiment was echoed across social media platforms, with the pseudonymous AI Leaks and News account noting an “overwhelmingly negative” consensus on both X (formerly Twitter) and Reddit. A primary source of user discontent stems from GPT-5’s new “router” feature, designed to automatically select a “thinking” or “non-thinking” mode based on query difficulty. Users report that this router frequently defaults to the less capable non-thinking mode, hindering performance for complex queries.

This lukewarm reception poses a significant challenge for OpenAI, especially as competition intensifies. Major rivals like Google and Anthropic are rapidly advancing their own large language models. Moreover, a growing array of powerful, often free and open-source, Chinese LLMs are emerging, offering competitive features. For instance, Alibaba’s Qwen 3 model recently updated its context window to 1 million tokens, allowing for nearly four times as much information exchange in a single interaction compared to GPT-5. OpenAI’s other recent release, the open-source GPT-OSS models, also received a mixed response. This broader context, coupled with a recent Polymarket betting market prediction favoring Google to have the best AI model by the end of August 2025, suggests a shifting landscape.

Despite the initial setbacks, some industry insiders, such as Otherside AI co-founder and CEO Matt Schumer, suggest that negative views may evolve as users and developers optimize their integration approaches for the new model. Schumer posits that there is typically a time lag between a new model’s release and when companies fully adapt their systems to leverage its capabilities. However, these early indications suggest that GPT-5 is not the “home run” release that previous iterations like GPT-4 or 4o were. This is a concerning signal for OpenAI, a company that, despite recently securing another funding round, remains unprofitable due to its extensive research and development costs.