Overview
- xAI is deploying two Grok 4.1 variants—standard and Thinking—across grok.com, X, and iOS/Android, with free access for all users and fewer usage limits for paying customers.
- xAI reports hallucination rates fell from about 12% to roughly 4% versus the prior Grok 4 Fast model, based on internal evaluations using real‑world queries and FActScore tests.
- Early crowdsourced results place Grok 4.1 (Thinking) at or near the top of LMArena/Text Arena leaderboards, though such rankings and company tests carry methodological caveats.
- A silent staged rollout from November 1–14 included blind pairwise comparisons on live traffic, where Grok 4.1 responses were preferred 64.78% over the previous production model.
- xAI highlights gains in speed, conversational and emotional intelligence, and creative writing, as Elon Musk separately says a much larger, real‑time multimodal Grok 5 is planned for early 2026.