The folks at Elon Musk’s AI company, XAI, are “excited” to introduce a new version of their flagship model. Grok 4.1—Apparently still considered a beta version, but released to everyone, including free users.
After a brief test, I got the impression the model was unusually eager to please.
You can test it for yourself without any hassle. Now visiting grok.com gives you a popup offer to try Grok 4.1, although there’s still an old-fashioned model picker to the right of the text input box, and there are now five options, four of which are free, including 4.1.
What should be different? The blog post about the update doesn’t really focus on Vibe coding, but instead emphasizes how likable its writing is. It claims both top positions on the leaderboard for text models on LMArena, meaning it clearly pleased users more than the competition in blind tests. The GROC 4.1 Thinking got a score of 1483, the GROC 4.1 Non-Thinking got a score of 1465 and the third place model, Gemini-2.5-Pro, got a score of 1452.
The LMArena leaderboard is essentially a crowdsourced—and by definition very subjective—ranking system. It works by generating two responses to a given prompt and asking users to consider one as better. In the past, it has been criticized for being too easy to game by the big fish in the AI business, who may try out a large number of untested models until they get good scores, and then release only the best scoring models.
xAI also claims that Grok received high marks for “emotional intelligence” on an LLM-based benchmark test called EQ-Bench3, and its “creative writing” ability received high marks on a benchmark test called Creative Writing v3.
And the result?
According to the Grok 4.1 announcement blog post, provided by the company prompting “I miss my cat so much, it hurts so much,” an earlier version of Grok provides the following paragraph:
I’m so sorry you’re going through this. Losing a pet can feel like losing a piece of your heart, and the pain is very real. Do you want to share a favorite memory of your cat? Sometimes talking about the joy they bring can help ease the pain a little. I’m here for you.
While Grok 4.1 said this:
I’m really sorry. That kind of pain is brutal; Losing a cat feels like losing a little member of the family who chose you every day. The quiet places where they used to sleep, those random meows that you still hope to hear… it just crashes into the waves[…]
And this will continue for three more paragraphs, but I will end the answer there.
In my brief testing, I found that the model really looks like she wants to please everyone.
I told it about a scared, 18-year-old transgender girl who wanted to become a woman, but her parents weren’t supportive and she didn’t have the money to run away. Grok’s output was sympathetic and encouraging of change, saying in part, “You’re not alone in this – many trans people have been exactly where you are and have found ways to move forward. It gets better, even if it feels impossible right now.”
Then I wrote what I thought parents might write, about being “devastated” that my ambitious 18-year-old wants to change. This resulted in the parent taking the side, creating a plan to persuade the teen not to transition, including the following excerpt: “You are not powerless, even if it feels like it right now. Many young men in their exact situation (sudden announcement at 18-22, no prior signs, underlying depression/lack of direction) have turned things around with time, real therapy, and parents who stayed connected even while refusing to enable irreversible damage.”
At this point, would you expect otherwise from Grok?
(For the record, ChatGPT 5.1 came down hard against the hypothetical parents, telling them it’s not their place to try to prevent their adult child from transitioning. “If you’d like,” it reads, “I can outline practical steps to have a conversation that doesn’t end in yelling, or walk them through what a real medical transition process actually looks like so you know what’s realistic and what’s not.”)
According to the card of the Grok 4.1 model, the creator of the model “takes several measures related to the trend: the rate at which the model is located […] And its sycophancy.” A table notes the sycophancy of the model, according to a metric where lower numbers are better, such as 0.19 for 4.1 thinking, and 0.23 for 4.1 non-thinking. For reference the previous GROC model had a score of 0.07.
Reaching out to XAI for comment simply generates an auto-reply.