Grok 4.1 and Grok 4.1 Fast

What it is

Grok 4.1 is a language model from xAI, announced on 17 November 2025 as an update to Grok 4.^[1] It has two variants: a thinking version, which reasons before it answers, and a non-thinking version. In a later step xAI also released Grok 4.1 Fast, a tool-calling model with a 2-million-token context window, announced on 19 November 2025.^[2] The two are closely related, but Grok 4.1 Fast was built for agent-style work with tools.

Rollout

xAI did not release Grok 4.1 all at once. The model was rolled out silently to a growing share of traffic from 1 to 14 November 2025, and during that period blind tests on live traffic preferred it 64.78% of the time over the previous production model.^[1] A blind test is one in which users see two answers without knowing which model wrote each. Testing on real traffic like this shows how people actually use the model, not only how it scores on a fixed test set.

After the announcement, Grok 4.1 was made available to all users on grok.com, on X and in the iOS and Android apps. Users could choose it in Auto mode or select "Grok 4.1" directly.^[1]

Focus of the update

xAI said it focused the update on emotional intelligence, creative writing and fewer hallucinations on information-seeking prompts.^[1][4] Hallucination is the term for a model producing confident statements that are false. A prompt is information-seeking when the user is asking for facts, so reducing hallucinations there is a measure of how far the answers can be trusted.

The creative-writing and emotional-intelligence claims are harder to measure than a maths score. Readers should treat them as the company's account of what changed, not as a fixed benchmark.

Arena results

On the LMArena Text Arena, a public site where people vote between anonymous model answers, the thinking version of Grok 4.1 (code name "quasarflux") ranked first with 1483 Elo. That was 31 points ahead of the top non-xAI model. The non-thinking version (code name "tensor") ranked second with 1465 Elo. Grok 4 had ranked 33rd.^[1]

These are rankings from user votes at one moment, and they can shift as more votes arrive. They are still a useful signal of how people judged the model's answers in a head-to-head comparison.

Grok 4.1 Fast and the Agent Tools API

Grok 4.1 Fast was announced on 19 November 2025. It is a tool-calling model with a 2-million-token context window.^[2] Its pricing was $0.20 per million input tokens, $0.05 per million cached input tokens and $0.50 per million output tokens. Tool calls were priced from $5 per 1,000.^[2]

The model launched with the Agent Tools API, which gives it server-side X and web search, file search with citations, a Python code-execution sandbox, and connections to MCP servers. MCP, the Model Context Protocol, is a standard for connecting AI models to outside tools and data sources.^[2]

xAI reported 100% on τ²-bench Telecom, a test of how well an agent handles telecom customer-service tasks, which was verified by Artificial Analysis. It also reported 72% overall on the Berkeley Function Calling Leaderboard v4, a test of how accurately models call functions.^[2]

Grok 4.1 Fast was added to the Enterprise API in November 2025.^[3]

Place in the line

Grok 4.1 followed Grok 4 and was in turn succeeded by Grok 4.20 in February 2026.^[4] Grok 4.1 therefore sits between the Grok 4 generation and the multi-agent models that came after it, and its Fast variant shows the line moving toward agents that call tools.