It’s All GPT-5 Now

Following the divisive response to the new GPT-5, we take a look at what some of its users are missing with the new rollout. How does the fresh new model compare, not just to its own prior versions, but also to its competitors?

OpenAI first showcased GPT-5’s upgraded coding and writing abilities with a simple game and a generated poem. It wouldn’t just consolidate all of OpenAI’s previous AI tools–like o3, o4, GPT-4o, GPT-4.5–but exceed all past capabilities, whilst also having the ability to defer simpler tasks to models less strenuous on their internal systems.

But this doesn’t seem to be a reflection of the opinions of their userbase today. ChatGPT’s users have flocked to its Reddit page of the same name to criticise and debate GPT-5’s current state.

Inflaming discussions further, OpenAI’s charts showcasing GPT-5’s capabilities had also been misleading, with its originally accurate numbers misrepresented in graphic form and leading users to think GPT-5 would be much more of an upgrade than expected. In response to backlash on an OpenAI team Ask-Me-Anything Reddit thread, Sam Altman noted the error and promised a revised chart.

Complaints have also surfaced regarding the AI’s new lack of ‘personality’, alongside an apparent increased rate of mistakes despite modifying user prompts. With more inputs required before an ideal output, people were hitting GPT-5’s rate limits even on paid plans, with free plans running out far quicker than before.

Most controversial of all has been OpenAI’s initial choice to cut off access to all of its prior models, with GPT-4o now temporarily reinstated, but only for paid users, when users once enjoyed the freedom to choose between their favourite models to fulfil specific functions, such as OpenAI o3 or o4 for stronger reasoning capabilities.

The AI tech world, moving at its high-intensity speeds, has already deprecated many of its older models. Microsoft’s Azure AI has already been through multiple iterations, and even OpenAI’s older GPT-3 models have been discontinued. So why the reaction?

Clear terms and timelines are a key component of software deprecation. Some of Google Vertex AI’s functions are slated to shut down both this year and the next, but with a clear deprecation date and final shutdown deadline months after. ChatGPT’s deprecation came as a sudden shock especially with how much it deviated from the company’s prior practices, where older GPT models would still be available for use for quite some time after.

The most vocal contingency of users, who flocked to social media platforms to state their displeasures, were power users of its chatbot features, who preferred GPT-4.5 and 4o’s more ‘humourous’ personalities, noting that communication with GPT-5 felt more impersonal and cold which then extended to its creative writing outputs.

For as long as genAI has existed, people have humanized and placed emotional weight on communicative models. Users on OpenAI’s community forums have grieved the deprecations of older models, from writing obituaries for pre-GPT-3.5 models to referring to the deprecation of text-davinci-003 in favour of new models as ‘killing [OpenAI’s] middle child’. A Reddit user wrote, “BRING BACK 4o GPT-5 is wearing the skin of my dead friend [sic]”, which Sam Altman himself replied to, noting the ‘evocative image’ and promising 4o support.

But a growing group of supporters have begun speaking up, pointing out that GPT-5’s more ‘cold’ tone would help to delineate its position as a professional aide, rather than a digital companion. With the increase in AI companionship applications and the heavily emotional response to the loss of 4o and 4.5, it’s likely OpenAI is working on creating distance between the AI and the user.

When asked on the above AMA Reddit thread about ChatGPT’s chatbot functions, OpenAI Head of Engineering Sulman Choudhry writes, “We do believe that the way we interact with ChatGPT should change from asking questions to something more suitable for doing work.”

Its functionality fits the bill as well, with strong capabilities in text and code review despite a sore lack of the ‘EQ’ that GPT-4.5 was touted for.

This mismatch between OpenAI’s intentions for and an individual user’s usage of ChatGPT has reached its breaking point. When ChatGPT was first released in 2022, it was introduced as a model ‘which interacts in a conversational way’, using the term ‘chatbot’ to refer to it in its intro article. It’s clear that OpenAI is attempting to break into the Agentic AI market with its new focus on APIs for enterprise use, but the many corporations listed in its OpenAI for Business page aren’t the ones complaining about GPT-5: it’s vocal individuals who ‘just want a fun LLM to chat with us about what to have for dinner!’, as one Reddit user emphasised in a post with 281 upvotes.

In Altman’s X post on the matter, he writes, “We really just need to get to a world with more per-user customization of model personality.” On that note, we also need a world with more transparency on AI deprecation and clearer user terms of use. With growing demands from large enterprises and individual users with different needs and use cases, more diverse solutions are still playing catch-up with AI growth.