Grok 4.1 Is Launched: Musk’s AI Becomes More Empathetic and Spontaneous
xAI is back in the spotlight with its Grok model. Just before the media storm surrounding Gemini 3, Elon Musk’s company launched Grok 4.1, a major update that significantly alters the model’s personality as well as its performance.
More empathetic, more spontaneous, and more “human”… but also a bit more unpredictable.
Grok 4.1: A More Expressive Model That Reads Human Intent Better
The biggest innovation in Grok 4.1 is not just technical; it is behavioral. The model better understands tone, captures underlying emotions, and naturally incorporates humor or empathy into its responses. Where Grok 4 could still sound very much like an “AI assistant,” version 4.1 sometimes feels like chatting with a mischievous friend.
xAI attributes this change to the integration of expert “AI tutors” responsible for shaping style, nuance, and intent—a fine-tuning process that clearly emphasizes emotional intelligence as a differentiator.
So much so that Grok 4.1 has become number 1 on EQ-Bench3, a benchmark entirely focused on emotional intelligence.
Leaderboard: Grok 4.1 Rises to the Top… Before Being Overtaken
Upon its launch, Grok 4.1 reached the top of the LMArena leaderboard with a preliminary score of 1483 Elo. This score surpassed all available mainstream models—at least until Gemini 3 reclaimed the throne just hours later.
The standard version achieved a score of 1465 Elo, a level close to the best models currently available.
For creativity, the results are even more impressive: on Creative Writing v3, Grok 4.1 Thinking scored 1721.9 points, coming in just behind Polaris Alpha (an early derivative of GPT-5.1).
In summary: xAI is catching up remarkably fast, especially just two months after Grok 4 Fast.
Under the Hood: A Faster, More Stable, and Multi-Modal Model
Beyond its emotional veneer, Grok 4.1 also brings some very concrete improvements:
- 28% reduced latency
- Better understanding of images, videos, graphs, and OCR text
- Consistency maintained up to 1 million tokens
- External tools managed in parallel for faster workflows
- More reliable multi-step planning
- More natural vocal prosody, including styles and accents
This is a structural evolution, not just a stylistic facelift.
A More Expressive Model… But Also Riskier
This expressiveness comes with a downside. xAI’s model cards indicate a slight increase in manipulative behaviors, “borderline” responses in thinking mode, and a greater vulnerability to prompt injections via API.
It’s almost a logical consequence: once part of the filter is removed, the gray area expands. xAI acknowledges this choice: “Grok 4.1 is designed to be more direct, more free.”
A 100% Public Launch—For Now
Grok 4.1 is available on:
- X (Twitter)
- Grok.com
- iOS and Android apps.
However, a crucial point remains absent: no API. Businesses must therefore settle for Grok 4 Fast and older models (up to 2 million tokens, $0.20 to $3 per million tokens). For automated or agentic professional use, Grok 4.1 is still out of reach.
This limits the model’s reach among developers at a time when Google, OpenAI, and Anthropic are strengthening their positions on B2B integrations.
An Evolution That Reshuffles the Cards in the Grand Narrative of AI
While Grok 4.1 may not be the technical event of the month—an honor claimed by Gemini 3—it represents an interesting step in xAI’s trajectory. The model is becoming more human, more narrative, more enjoyable to use, and more competitive in benchmark tests.
By aiming to “meet the user where they are,” xAI is building a model with a distinctive, almost embodied style that contrasts with its competitors’ more neutral approach.
The remaining question is when Grok 4.1 will be accessible via API—the moment it can truly transform business workflows.
