Grok 4.1

// Overview

Grok 4.1 is xAI's flagship chat-and-reasoning model, released on 17 November 2025 as the new default Grok across grok.com, X, and the iOS and Android apps. Rather than chasing raw exam scores, this release targets everyday usefulness: xAI tuned it for higher emotional intelligence, more natural creative writing, and far fewer made-up facts. It ships in two modes — a default fast mode (codename 'tensor') and a deeper 'Thinking' reasoning mode (codename 'quasarflux').

The headline change is reliability. xAI reports Grok 4.1's hallucination rate fell from 12.09% to 4.22%, and its FActScore error rate dropped from 9.89% to 2.97% — both roughly a threefold improvement over the prior generation. xAI ran a quiet two-week rollout (1-14 November 2025) in which blind A/B testers preferred Grok 4.1's answers 64.78% of the time, and both modes posted top placements on the LMArena Text Arena and EQ-Bench3 emotional-intelligence leaderboards at launch.

Grok 4.1 launched as a consumer model first: it was available free (with rate limits) and removed from limits on SuperGrok, but was not exposed through the standard xAI developer API at release. Developers were instead pointed to the separate, lower-cost Grok 4.1 Fast model and Agent Tools API two days later. Grok 4.1 carries a 256K-token context window and an ~8K-token-per-response ceiling, with a knowledge cutoff around November 2024.

// Benchmarks

1. Hallucination rate (lower is better)4.22%
2. FActScore error rate (lower is better)2.97%
3. Blind A/B preference vs. prior Grok64.78%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Strengths

• Sharply lower hallucination rate — xAI reports a drop from 12.09% to 4.22% versus the prior generation
• Improved factual accuracy on FActScore (error rate 9.89% to 2.97%)
• Strong emotional intelligence and empathy, topping the EQ-Bench3 roleplay leaderboard at launch
• More coherent, less formulaic creative and long-form writing
• Available free to all users (with rate limits) on web, X, and mobile
• Optional 'Thinking' mode for harder, multi-step reasoning tasks

// Best for

• Everyday conversational assistance and brainstorming where tone and empathy matter
• Creative and long-form writing — stories, scripts, marketing copy
• Drafting and editing where factual reliability is important
• Roleplay and emotionally-aware companion interactions
• Research and Q&A inside its 256K-token context window
• Real-time discussion of current topics surfaced through Grok on X

// Grok (flagship) — every version

The full lineage of the Grok (flagship) line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.