BAF NEWS

Grok 4.1 Release and Rollout

xAI launched Grok 4.1 on November 17, 2025. The company says it is now available across Grok.com, the X app, and iOS and Android apps.

Evidence shows the rollout is immediate and broad. xAI writes, “Grok 4.1 is now available to all users,” while external outlets confirm the same reach across web and mobile touchpoints. Additionally, reports note it ships as part of the standard experience rather than a limited beta.

Notably, availability spans both Auto mode and a direct model picker. That matters because users can explicitly choose the new release on day one. Consequently, adoption friction is low.

Model Configurations and Access

Grok 4.1 comes in two configurations: Non‑Thinking (fast) and Thinking (reasons before responding). The distinction aligns response speed with task complexity.

According to xAI’s documentation, users can let Auto mode select the configuration or manually pick “Grok 4.1.” Therefore, the interface supports both casual and power users. In practice, the toggle makes the system’s reasoning cost visible.

Moreover, the two‑mode design attempts to balance latency and depth. Non‑Thinking mode favors quick answers to straightforward prompts. Meanwhile, Thinking mode aims for better chain‑of‑thought performance on complex reasoning tasks.

Silent Rollout and User Preference Testing

Before the public launch, xAI ran a “silent rollout” from November 1 to 14, 2025. During this period, it continuously evaluated Grok 4.1 against the prior production model in blind pairwise tests.

The reported preference rate was 64.78% in favor of Grok 4.1. In other words, users preferred responses from the new model in nearly two out of three comparisons. Importantly, this testing occurred across the same product surfaces it now powers.

However, these are internal evaluations disclosed by the company and echoed by media. As a result, they provide directional evidence rather than an independent audit. Still, the breadth of the stealth test suggests production-level confidence.

Leaderboard Performance

At release, xAI highlighted top-tier placement on LMArena’s Text Arena. As of the leaderboard’s November 16 update, Grok 4.1 Thinking sits around 1484 Elo and Grok 4.1 (Non‑Thinking) around 1465 Elo, placing both variants near the top.

The contradiction: xAI’s launch note touts a number-one moment, yet the public board shows “near the top” at its latest update. This is not unusual, since Elo fluctuates as new matchups stream in. Nevertheless, both configurations cluster with leading proprietary models.

Furthermore, Elo here reflects head-to-head user judgments. Therefore, it complements xAI’s internal evaluations with a competitive, platform-wide signal. The combined picture suggests early strength in real-world prompting contexts.

Reduced Hallucination and Improved Factuality

xAI also reports fewer hallucinations on information-seeking prompts and improved FActScore versus Grok 4. The company positions this as a major reliability upgrade.

Media coverage cites specific deltas from xAI’s internal graphs. One report describes “hallucination rates plummeting from 12.09% to 4.22% … errors dropped from 9.89% to under 3%.” According to available reports, these are internal measures rather than third-party audits.

However, the direction is consistent across write-ups and company materials. Moreover, the metrics target the failure modes users notice most: made-up facts and brittle citations. If sustained, these gains reduce friction for research and reporting workflows.

Emotional Intelligence and Creative Writing Improvements

Beyond factuality, xAI emphasizes better emotional and creative performance. It highlights gains on EQ‑Bench3 and Creative Writing v3 and calls the model “exceptionally capable in creative, emotional, and collaborative interactions.”

Consequently, Grok 4.1 may feel more aligned with nuanced intent. Moreover, improvements in style coherence and persona consistency can help long multi-turn sessions. This is particularly relevant to brainstorming, coaching, and editorial assistance.

Still, these are company-reported benchmarks rather than multi-lab studies. Therefore, external replication will be important. Yet early examples point to more grounded tone and fewer abrupt shifts.

Platform Accessibility Enhancements

Coverage notes that users can access Grok 4.1 on the website without logging in. That lowers barriers for first-time trials and casual queries.

In addition, the model remains integrated within the X app and the mobile apps. Therefore, the touchpoints cover both public web and authenticated environments. For distribution, this is a straightforward way to expand reach quickly.

Finally, pairing easy access with Auto mode reduces setup time. New users can sample the model’s default behavior without navigating settings. In turn, adoption likely benefits from fewer taps and fewer hurdles.

Evidence Recap

Evidence shows a rapid release, two configurable modes, and a strong user-test preference. It also shows high placement on a public leaderboard and improved factuality metrics.

However, many results come from xAI’s internal evaluations. Therefore, continued third-party testing will matter. Even so, the early signal is unambiguous: Grok 4.1 is launching with momentum.

Quotes Referenced

• “Grok 4.1 is now available to all users.”
• “Grok 4.1 is available in two configurations: Grok 4.1 Non‑Thinking … and Grok 4.1 Thinking.”
• “Silent Rollout, November 1–14, 2025.”
• “Grok 4.1 is preferred 64.78% of the time.”
• “hallucination rates plummeting from 12.09% to 4.22% … errors dropped from 9.89% to under 3%.”
• “Users do not even need to be logged into their accounts if they access the model on the website.”