Grok (xAI)

Overview

xAI, founded by Elon Musk in 2023, develops the Grok series, distinguished by real-time access to X (Twitter) data and lightly filtered responses. As of August 2026 the flagship is Grok 4.6 (released August 12, 2026): a 500K-token context model tuned for long-running agents, agentic coding, and interactive or visual work. It is also available through Amazon Bedrock.

Current model: Grok 4.6

Lineup

Model	Context	Input	Output	Notes
Grok 4.6	500K	$2/M	$6/M	Current flagship, built for long-running agents; also on Amazon Bedrock
Grok 4.1 Fast	2M	$0.20/M	$0.50/M	Very low cost, largest context

Past 200K tokens, Grok 4.6 switches to a long-context band of $4 input / $12 output — and xAI applies the higher rate to every token in that request, so budget for double if you plan to fill the 500K window.

Key capabilities

1. 2M-token context (among the largest)

Grok 4.1 Fast supports a 2,000,000 token context, the longest of any Western closed model.

2. Very low cost

Grok 4.1 Fast at $0.20 / $0.50 per 1M tokens sits in the same bracket as GPT-5.6 Luna ($0.20/$1.20) and Gemini 3.7 Flash ($0.75 input), making it one of the best value reasoning models available.

3. Tuned for agentic coding

Grok 4.6 posts large gains over the previous generation on agent benchmarks such as DeepSWE and APEX-Agents, and long-running coding agents are its stated target workload.

4. Real-time X integration

The only major LLM that can reference live posts and trends on X, so answers reflect what is happening right now.

5. DeepSearch

Multi-source research, available even on the free tier, combining live X data with web search.

6. Grok Build (coding agent)

A terminal coding agent supporting AGENTS.md, plugins and skills, hooks, and MCP, with a plan mode and parallel sub-agents. xAI open-sourced it on GitHub in July 2026.

7. Image and video generation

Imagine Image 2.0 for stills and the grok-imagine-video models for short clips from a still image are both reachable inside Grok.

Plans

Plan	Monthly	Highlights
X (free)	$0	Basic Grok, DeepSearch
X Premium	$8	Extended Grok access
X Premium+	$40	Ad-free X plus high Grok throughput; 50% off SuperGrok
SuperGrok	$30	Newest models, DeepSearch, Grok Build
SuperGrok Heavy	$300	Multi-agent Heavy mode, the largest allowance, priority access, early product previews
xAI API	Pay-per-use	Developer API

If Grok itself is the goal, SuperGrok at $30 needs no X account and delivers more usage per dollar than X Premium+ at $40.

Typical uses

• Real-time information: X trends and breaking news
• Long-context work: whole books or codebases via the 2M window
• Cheap reasoning: Grok 4.1 Fast for high-volume inference
• Agentic coding: long-running tasks through Grok Build
• Content generation: informal writing and social posts

Strengths

• ✅ 2M-token context (Grok 4.1 Fast, among the largest)
• ✅ Grok 4.1 Fast is one of the cheapest reasoning models
• ✅ Real-time X data access (unique)
• ✅ Grok Build is open source and included from SuperGrok ($30) up
• ✅ Always-on reasoning
• ✅ Minimally filtered responses

Weaknesses

• ❌ Limited Japanese support (English and code first)
• ❌ Immature ecosystem (plugins and marketplace still developing)
• ❌ Grok 4.6 doubles its rate above 200K tokens, making long-context spend hard to predict
• ❌ Heavy dependence on X — outside the platform, access is API-only
• ❌ Relaxed safety filters — worth care in enterprise deployments

Official resources

• Website: https://x.ai
• Grok: https://grok.com
• API docs: https://docs.x.ai
• Pricing: https://docs.x.ai/docs/models