GPT-5 reveals frustrations over AI model progress

Analysts have advised users to lower their expectations for upcoming AI models, but investors remain confident in the artificial intelligence industry. The tempered sentiments came after the launch of OpenAI’s GPT-5 disappointed users and analysts with only minor improvements. 

Expectations were high when OpenAI unveiled its GPT-5 model last week. The company’s Chief Executive, Sam Altman, described it as “a significant step along the path to AGI,” its acronym for Artificial General Intelligence. Executives also believed this release would address weaknesses in ChatGPT.

However, the reception was not as expected. Users quickly posted screenshots of GPT-5 failing simple tasks on social media, from mislabeling U.S. maps to stumbling over basic arithmetic.

GPT-5 reveals frustrations over AI model progress

Advanced users criticized GPT-5’s personality, describing the chatbot as colder and more mechanical than its predecessors. OpenAI’s update acknowledged the issue, promising to make the model warmer and friendlier, while emphasizing that sycophancy levels had not increased compared to previous models.

The new usage limits, initially capped at 200 queries per week, also got paying subscribers angry. Altman was forced to take back some decisions, reinstating a popular earlier model that had been retired, and promised more customization options.

The disappointment pointed to deeper frustrations with the pace of AI advancement. Previous releases were seen as leaps in capability, but GPT-5 has been described as incremental. “For GPT-5, people expected to discover something totally new,” said Thomas Wolf, the co-founder of  AI startup Hugging Face. “And here we didn’t really have that.”

Compared to competitors such as Claude, Gemini, DeepSeek, and xAI, GPT-5’s performance was consistently mid-level across scientific, coding, and customer service tasks, according to research from Princeton University. The model’s key strengths were cost efficiency and speed, not outstanding intelligence.

This has brought back comparisons to the “AI winter” of the 1980s, when inflated expectations collapsed as companies failed to deliver commercially viable systems. Gary Marcus, a prominent critic, said, “GPT-5 was this central icon of the entire approach of scaling to get to AGI, and it didn’t work.”

For years, progress in large language models has depended on pouring more data and computing power into training systems. Companies have now exhausted much of the freely accessible internet data, forcing deals with publishers and rights holders, while the costs of training keep soaring. GPT-5 reportedly used hundreds of thousands of Nvidia’s latest processors, highlighting the scarcity and cost of resources.

Some observers argue that the fixation on scaling has restricted exploration of alternative approaches. Meta’s chief scientist, Yann LeCun, suggested that progress requires models trained on real-world video and multimodal data.

Investors double down despite slowing AI hype

In all this, investors have remained mostly unfazed. Wall Street and venture capital have continued to pump billions into AI.

Nvidia, the dominant chipmaker powering most AI training, has seen its valuation rise to $4.4T. Shares in SoftBank, a major OpenAI supporter, have surged more than 50% in the past month. At OpenAI itself, annual recurring revenue from ChatGPT has reached $12B.

The confidence of investors is in contrast to expectations from analysts. Princeton researcher Sayash Kapoor noted that GPT-5 was less about approaching AGI and more about serving as infrastructure for products. The model’s reliability and cost-effectiveness could encourage businesses to build new applications on top of it.

Venture capitalists also argue that opportunities remain underexplored. “Start-ups and businesses have not begun to scratch the surface of what they are capable of in business and consumer applications,” Peter Deng, a general partner at Felicis and former OpenAI executive, said.

Get seen where it counts. Advertise in Cryptopolitan Research and reach crypto’s sharpest investors and builders.