Grok 4.6 (Fable) by XAI: a new AI benchmark, plus Grok bot f

Summary

XAI has released Grok 4.6 (Fable), positioning it as a serious OpenAI and Anthropic competitor on both capability and cost. The model is built for advanced reasoning and high-quality engineering data, and it arrives alongside Grok bot, a cloud-based agent platform designed for continuous digital labour. Here is what Grok 4.6 is, why it matters, and how to access it.

What is Grok 4.6 (Fable)?

Grok 4.6 (Fable) is XAI's new 1.5 trillion parameter base model, designed to improve reasoning and technical understanding. It is also aimed at producing high-quality engineering data for complex professional tasks.

According to the release coverage, Grok 4.6 has been benchmarked alongside top-tier models including Fable 5 Max, OpenAI's GPT-4, and Anthropic's Claude, with results that suggest it can handle intricate, domain-specific work. The improvement over Grok 4.5 is attributed to a longer supplemental training run, using expertly curated data from Cursor, plus a refined training recipe.

Why Grok 4.6 stands out

The headline is not just benchmark performance, but practical output quality in real-world tests. The coverage highlights examples where Grok 4.6 can generate detailed game clones such as Portal 2, including physics and 3D models, in a single run.

It also performs strongly on GPT Val benchmarks, which the release frames as evidence of its ability to tackle complex, industry-specific challenges. Beyond raw capability, the model is described as responsive and clear, making it well suited to day-to-day AI collaboration.

How Grok 4.6 and Grok bot work together

Grok bot is presented as a cloud-based AI agent platform for digital labour. Unlike tools that require constant user presence, it runs continuously via virtual machines, so it can keep working without the user's device being online.

Key Grok bot features described include:

Creating custom AI agents
Assigning teachable tasks to reduce risks such as unintended messaging
Using a Grok bot interface with multiple-choice style prompts to streamline decisions and reduce cognitive load

The release also notes that Grok 4.6 may not yet be live on Grok bot, but XAI is expected to deploy it soon. The plan includes continuous monitoring by dedicated AI agents to support timely updates, and to improve delegation and management through sub-agents.

When and where to access Grok 4.6

Grok 4.6 is available now. The launch week includes promotions that offer double usage on Cursor and Grok build platforms.

Pricing is positioned as highly competitive, starting at:

$2 per million input tokens
$6 per million output tokens

The coverage frames this as about half the cost of similar models from OpenAI and Anthropic, which may appeal to organisations looking for advanced capability without matching top-tier pricing.

XAI's roadmap: what comes next

XAI's development pace is part of the story. The release points to accelerated training and deployment efforts drawing on data from the SpaceX acquisition and partnerships with Cursor.

The roadmap includes:

Grok 4.7, a 2.1 trillion parameter model, expected in just a few weeks
Grok 5, targeted for release by year-end

Frequently Asked Questions

What is Grok 4.6 (Fable)?

Grok 4.6 (Fable) is XAI's 1.5 trillion parameter AI model, built for advanced reasoning, technical concept understanding, and high-quality engineering data.

Why is Grok 4.6 important compared to other AI models?

The release claims it matches or exceeds performance from leading models such as OpenAI and Anthropic, while also offering strong real-world results and lower pricing.