xAI has officially released Grok4.1 today, claiming a 42% reduction in response latency, an 18% improvement in intent recognition accuracy, and significant optimization in dialogue coherence. The new version continues to be based on the Grok-4MoE architecture, achieving a "second-to-second" experience by adding a real-time feedback layer and personalized cache. It remains available unlimited times for X Premium+ users, with API pricing remaining at $5 per million tokens.

Internal tests show that Grok4.1 has set new records in multiple benchmarks: MT-Bench score of 8.97, breaking 8.9 for the first time; HumanEval code generation pass rate remains at 87.1%; multi-turn dialogue consistency reaches 91.4%, an increase of 6 percentage points from the previous version. Additionally, the model now includes a "context memory" switch, allowing users to choose whether to retain interaction records within the last 30 days to optimize response style.

An xAI product representative revealed that Grok4.1 will serve as a preview for the upcoming "Grok5" at the end of the year, focusing on verifying the reinforcement learning plus human feedback (RLHF) mechanism. The model is now fully deployed to xAI's cloud, and the web version and X App have been updated simultaneously, with mobile updates coming later.