xAI has begun providing its new model, Grok 4.1, to all users via grok.com, X, and its iOS and Android apps. The model is being rolled out immediately in Auto mode and can also be explicitly selected via the model picker.
The company explained that 4.1 features significant improvements in creativity and emotional interaction. It utilized the same large-scale reinforcement learning infrastructure as Grok 4 to optimize style, personality, and utility. xAI stated that it developed a new method utilizing frontier agent reasoning models as reward models for reward signals that cannot be verified.
During a two-week silent rollout, blind pairwise evaluations were conducted using live traffic. Grok 4.1 was preferred at a rate of 64.78% compared to the previous generation model.
In LMArena's Text Arena, Grok 4.1 in reasoning mode ranks first overall with an Elo of 1483. The non-reasoning mode also secured second place with an Elo of 1465, surpassing the full reasoning settings of other models.
Additionally, performance was measured using benchmarks such as EQ-Bench3 and Creative Writing v3. As an example of a response to an emotional prompt, the model demonstrated a polite dialogue empathizing with the loss of a cat.
Source: Grok 4.1 (HN 140pt, 128 comments) (HN Search (backfill), 2025-11-18)