Here are the key improvements in Grok 4.1 as compared to its previous models:

▸ Better user preference: In blind pairwise tests during rollout, Grok 4.1 was preferred ~64.78% of the time over the previous production model.

▸ Enhanced emotional and interpersonal ability: It performs stronger on emotional-intelligence benchmarks (e.g., EQ-Bench) and is more capable at nuanced, empathetic responses.

▸ Improved creative writing and style: In benchmarks for creative writing, it shows more engaging, coherent personality and better stylistic fluency.

▸ Reduced factual errors (hallucinations): Post-training emphasis was placed on real-world information-seeking prompts; the hallucination rate on sampled production queries went down.

▸ Maintains strong reasoning and general capability: While improving style and interaction, it retains the “razor-sharp intelligence and reliability” of its predecessors.

▸ Better alignment of style/personality/helpfulness: Training methods were updated to more deeply optimise non-verifiable reward signals (style, alignment, personality) using advanced agent-based reasoning models as reward models.

点击图片查看原图