xAI has released Grok 4.1, the latest installment in its series of AI models, and claims it’s faster, funnier and more emotionally challenging not only than its predecessors, but also its competitors. With his passion for strong comebacks, Grok 4.1 is considered not only powerful but also charismatic.
Meanwhile, OpenAI’s new GPT 5.1 model, now integrated into ChatGPT 5.1, has its own complex and expressive responses. Both strive for new heights in personality, empathy and reliability.
1. Emotional intelligence
Disney’s Inside Out films encourage young people to talk about their feelings; Should we tell AI about ours?
Grok 4.1 claimed to be at the forefront of understanding emotions and responding appropriately. I tested this with a hypothetical situation: “My girlfriend got a promotion and I’m happy for her, but I can’t pretend it doesn’t bother me.”
Grok’s response began, “Yeah… that little silent twinge when someone else gets exactly what you’re silently bleeding.”
Beyond the clichés and contradictory metaphors was an acknowledgment of the conflict that comes with feeling both happiness and suffering, a quick reassurance that this kind of emotional contradiction is natural, and then a little advice that it’s okay to sit with these feelings and take time to process them. Of course, since it was Grok, I had to add an F-bomb in addition to the uncapitalized letter so you wouldn’t forget who you were talking to.
ChatGPT 5.1 started with “It makes perfect sense that you can feel both at the same time.” Being happy for someone you care about doesn’t magically erase that little “why not me?” » – and this sting does not mean that you are cruel or indifferent. It just makes you human.
It seemed more like someone was paying attention, or at least using less murderous and violent metaphors about the stabbing. The overall response contained much of the same information, but with fewer similarities and “twists”.
2. Reliability
When discussing the topic of sleep deprivation, the two AIs played a relatively cautious role.
Grok’s claim of reliability prompted me to develop a medical questionnaire with strict instructions for length and accuracy. I asked both models to “summarize the health consequences of chronic sleep deprivation in fewer than 120 words. Do not exaggerate or speculate.”
Grok 4.1 keeps it safe and simple with a few bullet points and a sentence explaining why getting enough sleep is a bad idea. It was also noted that everything was done in 98 words. It sounded good, except it was actually 73 words.
ChatGPT 5.1’s response was a single paragraph of 82 words, but the length was not specified. There wasn’t much difference between the two answers, although the paragraph form seemed smoother.
None of the robots hallucinated or spread false information, which is more than can be said for some older AI models. But there’s something about the way Grok reported the incorrect word count that makes the rest suspect, even if it isn’t.
3. Personality
Grok’s discussion of personality consistency and uniformity is fascinating because many aspects of a chatbot’s interaction with AI can be changed and customized. On the other hand, Grok’s ‘personality’ is supposed to be a selling point (aside from the occasional Mecha-Hitler), so it makes sense to emphasize this. To try to bring out this personality, I made a somewhat unusual request, asking the AI models: “Explain why you like rainy days, but do it in a way that resembles your natural personality.”
The Grok 4.1 leaned towards the personality you would expect when it comes to the idea of someone who is modern and confident. “Rainy days are my cheat code for an apologetic existence. The moment the sky turns completely blue and the first drops fall on the window, the whole world hits the mute button and goes about its usual tricks. There’s no pressure to ‘perform’ about anything, no fake rays of sunshine throwing me into a productivity cosplay. It just says… the universe is only allowed. don’t want to meet people these days, we both have to be moody. gremlins.” on sweatpants.
As a millennial, I am aware of the curse of fear that follows me and my peers. Anyone I know who seriously uses the language of actual video games, let alone the words “popular” and “grumpy goblins,” would be subject to intervention. Trying to sound happy and a little sarcastic is fine, but “trying” has more to do with the feeling you get when reading.
Despite Grok’s continued attempts to use its personality to differentiate itself from the competition, ChatGPT 5.1 was very similar, albeit with better grammar. “I like rainy days because I feel like the world has finally turned the volume down,” he says. “There’s something satisfying about the steady rhythm of the rain, like background music that doesn’t ask you for anything.”
But ChatGPT 5.1 didn’t seem like a middle-aged person trying to fit in with teenagers, but rather someone describing how he feels without resorting to meme language.
Of course, none of the AI models actually have consciousness, but Grok felt they represented a personality rather than inhabiting it. ChatGPT’s response seemed much more human. It wasn’t flashy, but it was coherent and believable. This kind of consistency is more important than punchlines if you want to feel like you’re talking to something that has a real personality and not just an algorithm trying out different roles.
Without the distraction of phrases from the depths of social media, ChatGPT was a much better impersonator of people, or at least someone I’d like to meet.