Why GPT-5 Feels Like a Step Forward and a Step Back

When better performance challenges user expectations

Image created with ChatGPT

OpenAI released GPT-5 on August 7, amid sky-high expectations. CEO Sam Altman described it as “a significant step on the path toward artificial general intelligence.” However, he also acknowledged that GPT-5 still lacks key capabilities, such as continuous learning.

In his usual ambitious style, Altman claimed that, for the first time, it truly feels like you’re talking to an expert on any subject—someone with the depth and insight of a PhD.

This messaging raised expectations among users, many of whom were hoping for a groundbreaking upgrade. But just a few days after launch, the reaction from the community made one thing clear. GPT-5 has sparked mixed feelings. While some users and experts have praised its new features and improvements, others have been deeply disappointed, calling the release underwhelming and far from the revolutionary leap they were hoping for.

Ultimately, opinions on GPT-5 depend on who you ask.

The Disappointed: "Overhyped and Underwhelming"

Despite all the hype, a group of users and analysts felt that GPT-5 fell short. AI critic Gary Marcus was especially vocal, calling the release “just another small update” that seemed rushed.

Marcus argued that OpenAI was overstating the impact of a model that, at its core, remains fundamentally similar to its earlier versions.

Early benchmark results supported the concerns of skeptics. GPT-5 scored just 56.7% on the SimpleBench evaluation, ranking fifth overall and performing worse than some smaller models. This result led to speculation that OpenAI may have prioritized scalability and profitability over significant improvements in performance.

When it came to basic reasoning tasks, testers observed only minor progress, if any, compared to GPT-4. Math puzzles, logic problems, and common-sense questions that confused GPT-4 continued to stump GPT-5 unless given extra attention and structure.

Factual accuracy also remains a challenge. GPT-5 still tends to hallucinate at times and can miscount letters or invent details without clear and careful prompting from the user.

Beyond the technical metrics, user feedback was particularly harsh. For a segment of longtime ChatGPT users, GPT-5 felt like a step backward. A common complaint was the loss of the model’s previous “personality.” Many users felt that while GPT-4, especially the 4o version, came across as friendly, creative, and even witty and warm, GPT-5’s responses felt dry and overly brief.

“The answers are bland and unhelpful. These are the most uninspired replies I’ve seen,” wrote one Reddit user, comparing GPT-5 to “an HR employee at the end of a long day.”

This user, along with others, expressed disappointment that what once felt like a conversation with a thoughtful and approachable AI had become mechanical and predictable.

Criticism also extended to functionality and user experience. GPT-5 was initially rolled out as a single unified model, with no option to switch back to earlier versions. This upset users who preferred older models for specific use cases. The sudden removal of GPT-4o and other variants felt to many like a forced transition rather than an upgrade.

In addition to losing access to familiar models overnight, users also discovered that GPT-5 came with stricter usage limits. Many quickly reached their message caps and were locked out of further use for several hours.

Given the strong backlash, OpenAI responded quickly. The company admitted that GPT-5 had not delivered the fully unified solution it had promised. In response to growing concerns, CEO Sam Altman hosted a Reddit Q&A to address issues with the new GPT-5 routing system, which users said often led to inconsistent or subpar responses.

Perhaps the most telling move came just a week after launch, when OpenAI reversed a key design decision. The model selector returned to the ChatGPT interface. Instead of forcing users into a single mode, OpenAI introduced new options labeled Automatic, Fast, and Thoughtful for GPT-5. Paid users also regained access to older models like GPT-4o, GPT-4.1, and o3. GPT-4o was even restored as the default model, offering fans of its more expressive tone a familiar choice.

This shift represented more than just a technical adjustment. It reflected an acknowledgment that there is no one-size-fits-all model for everyone. The simplified interface Altman had originally hoped to streamline turned out to be limiting for many users who valued flexibility and control.

The attachment users feel toward specific AI personalities is more than just a preference. It has become a key factor in how people interact with and adopt these tools. OpenAI’s rapid course correction, including the return of older models and a renewed focus on customizable AI personalities, highlights just how important user sentiment has become in shaping the direction of AI development.

The Enthusiasts: Finding GPT‑5's Silver Lining

While some users expressed disappointment with GPT-5, many others have focused on its strengths and the new opportunities it introduces. These users and experts view GPT-5 as a meaningful step forward. OpenAI describes it as “our most intelligent, fastest, and most useful model yet,” and there is substance behind that statement.

One of the most notable changes is GPT-5’s unified architecture. This design combines multiple sub-models with an intelligent routing system that decides when to respond quickly and when to take more time for deeper, more thoughtful reasoning. As a result, users no longer have to manually choose between a faster but less thorough model and a slower, more detailed one. GPT-5 makes that decision automatically and can even pick up on prompts like “think carefully about this” to adjust its response style.

In theory, this approach provides the best of both worlds. Users get fast answers for straightforward questions and more analytical responses when the task is complex.

“I’ve really enjoyed GPT-5 so far; it’s much better for me than 4o,” another Reddit user commented. They pointed out that the extended reasoning mode shows “surprisingly strong writing abilities,” while the standard mode still produces creative output that feels similar to GPT-4o when needed. For users like this, GPT-5 offers a better overall balance. It handles everyday tasks efficiently while applying deeper reasoning to more challenging problems.

Nick Turley, ChatGPT’s Director at OpenAI, noted that “this model is really appealing” to the average person who simply wants helpful answers. While that might sound like typical PR, it resonates with users who found the conversational style of GPT-4o to be overly elaborate for their needs.

Beyond usability, GPT-5 introduces several technical upgrades that have generated genuine excitement. One of the most significant is its expanded context window, which allows GPT-5 to manage extremely long documents or extended conversations without losing track of context.

Businesses, in particular, have responded positively to GPT-5. On launch day, OpenAI highlighted that companies such as BNY Mellon, SoftBank, Lowe’s, and Morgan Stanley have started equipping employees with tools powered by GPT-5.

Organizations are recognizing the value of the unified model, especially when managing multi-step workflows. Early feedback from enterprise users has been promising. For example, the head of AI at Amgen reported that GPT-5 met their “highest standard for scientific accuracy.” The model was able to answer complex research questions more effectively than previous versions and provided “higher quality and faster results” during internal testing.

What makes this especially promising is that GPT-5 goes beyond basic chat functions. It is increasingly being integrated into company systems and workflows to support coding, customer support, data analysis, content generation, and more.

Striking a Balance: What GPT-5's Reception Tells Us