- Sam Altman says GPT-5.4 is his “favorite model to talk to” and a step forward for ChatGPT’s personality
- OpenAI admits that the model still has three notable weaknesses, including design taste, real-world context, and task completion
- The comments suggest that OpenAI is now focusing as much on how ChatGPT feels to use as how powerful it is
In a recent post on X, OpenAI CEO Sam Altman described the new ChatGPT-5.4 as “my favorite model to talk to!” while also admitting that previous fifth-generation models had not been perfect when it came to human interaction. “We’ve been missing the mark on the model personality for a while, so it feels extra good to be moving in the right direction,” he added.
His use of the phrase “for a while” appears to refer to the debate over ChatGPT’s personality that has dogged OpenAI ever since it retired the popular ChatGPT-4o model last November. Many users felt that the 4o had a much better personality than the later fifth generation GPT models.
Campaigns to bring back the 4o model have been running since it was removed on 13 February 2026, despite access to fifth-generation GPT models being widely available.
The article continues below
OpenAI has recently released new GPT-5.3 Instant and GPT-5.4 Thinking models, and it is this 5.4 model that Altman believes has the best personality as well as being good at coding.
GPT-5.4 is great for coding, knowledge work, computing, etc. and it’s nice to see how much people are enjoying it. But it’s also my favorite model to talk to! We’ve been missing the mark on the model personality for a while, so it feels extra good to be moving in the right direction.March 7, 2026
The best model in the world
However, Shumer went on to list three weaknesses. Specifically:
- “Frontend flavor is way behind Opus 4.6 and Gemini 3.1 Pro. Why is this so hard to fix? @OpenAI once you fix this there is literally no reason for me to use any other model. Please do it!”
- It can still miss obvious context in the real world. For example, I had to plan an itinerary for a trip. At first glance it looked perfect, but it didn’t take into account that it was picking locations that would be bullied by spring breakers, so I had to re-run the prompt from scratch with more context.
- When testing it inside OpenClaw, it kept crashing before completing tasks. I assume this will be fixed quickly, but it’s still worth noting.”
In response, Altman enthusiastically wrote, “We will be able to fix these three things!”
You might be wondering what “frontend flavor” means here. We think Shumer is referring to AI’s sense of style and aesthetics when generating user interfaces for things like web apps, so this is related to coding.
While he praises GPT-5.4’s coding prowess, it sounds like the user interfaces it produces are less polished, modern or visually appealing than those generated by rivals like Claude or Gemini.
The other two criticisms seem more straightforward, although if you’re wondering what OpenClaw is, it’s an open source system designed to automate AI workloads across large numbers of Macs, particularly clusters of Mac minis. It can be used to combine multiple Macs to run or train AI models.
The popularity of OpenClaw is blamed on shortages in the Mac mini supply chain.
More than anything, the exchange suggests that OpenAI is now focused on how ChatGPT feels to use, not just how well it works on paper. That’s significant because many of the complaints about fifth-generation GPT models weren’t really about capability, but about personality – with people criticizing them for sounding robotic, flat or overly sycophantic.
Altman’s post suggests that OpenAI knows that getting the tone right is no longer a side issue, but an important part of making a GPT that people will actually talk to.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews and opinions in your feeds. Be sure to click the Follow button!
And of course you can too follow TechRadar on TikTok for news, reviews, video unboxings, and get regular updates from us on WhatsApp also.
