Grok

Grok is SpaceXAI's flagship AI assistant, distinguished by its native, real-time access to information from the X platform. Unlike most assistants that rely on static training data, Grok can reason about what's happening right now — making it the preferred tool for power users who need current intelligence alongside deep reasoning.

Overview

Launched in November 2023 by xAI, Grok has evolved into a frontier-class AI system powered by the Colossus supercluster. The assistant is built with a "maximally truthful and curious" design philosophy, delivering direct, unfiltered answers on complex and sensitive topics that other models often avoid.

Grok is no longer an xAI product. SpaceX acquired xAI in an all-stock transaction that closed on February 2, 2026, structuring it as a wholly owned subsidiary. In July 2026 the unit was rebranded SpaceXAI, and xAI ceased to exist as a standalone company. Grok and X are now SpaceX's AI division. The product itself did not move: grok.com and the X integration work as before.

The assistant offers a standard chat variant and a built-in Reasoning mode for deep, multi-step logical problem-solving. It is deeply integrated into the X ecosystem but also available as a standalone product and via API for developers.

The model roster and the app are documented separately, and only one of them is documented well. Grok 4.6 shipped on August 12, 2026 — five weeks after Grok 4.5 — and, like 4.5 before it, arrived as a developer-facing release: SpaceXAI's announcement names the API, Grok Build, Cursor and partners such as OpenRouter, Vercel and Cloudflare, and does not mention the consumer Grok app or the X integration. SpaceXAI does not publish which model each consumer tier is served, so this page describes the API roster where it is certain and says so where it is not.

Two things are worth knowing before you assume 4.6 is a straight upgrade. It carries a 500K context window, half of Grok 4.3's 1M, and its output price is $6.00 per million against Grok 4.3's $2.50 — so for whole-corpus prompts the older, cheaper model is still the right tool. And there is no Grok 4.6 Fast: the "faster variant at double the price" reported at launch is Priority Processing, a service tier rather than a model.

Key Features

• Real-Time X Integration: Native access to the live global conversation on X, enabling real-time news analysis, trend monitoring, and event summarization unavailable to other models.
• Reasoning Mode: A "think-before-you-respond" mode that allocates extended compute to solve complex math, code, and logic problems. On the API, Grok 4.6 exposes four effort levels — low, medium, high and xhigh.
• DeepSearch: An autonomous web research agent that synthesizes information from multiple live sources into a single, cited answer.
• Multi-Modal Understanding: Native vision capabilities for analyzing images, charts, documents, and screenshots.
• Grok Imagine: Built-in image and video generation, exposed to developers through the Imagine API.
• Grok Build: A separate coding model (grok-build-0.1) trained for agentic coding workflows, with a 256K-token context window.
• Voice Mode: Real-time, low-latency voice conversations, plus a Voice Agent API.
• API Access: OpenAI-compatible REST API for developers building on top of the Grok model family.

How It Works

Grok processes your requests through several specialized systems working in parallel:

1. Live Data Retrieval: Simultaneously queries X's firehose and public web sources for real-time context.
2. Reasoning Engine: Allocates thinking tokens to plan and verify multi-step answers before generating output.
3. Response Generation: Produces a direct, cited response using the combined context of live data and model knowledge.

Technical Architecture:

• Models: Grok 4.6 (the current flagship), Grok 4.5 and Grok 4.3 all remain callable. Grok Build 0.1 is the separate coding model.
• Context Window: 500,000 tokens for Grok 4.6 and 4.5; 1,000,000 for Grok 4.3; 256K for Grok Build 0.1.
• Reasoning: Configurable — low, medium, high and xhigh on Grok 4.6.
• Training Infrastructure: xAI's Colossus supercluster.
• API Standard: OpenAI-compatible REST API.
• Multimodality: Text, Image (Vision + Grok Imagine generation), Voice.

Use Cases

Research & Information

• Real-Time News Analysis: Tracking breaking events and synthesizing multiple live sources into a coherent summary.
• Market Intelligence: Monitoring social sentiment, trend shifts, and competitive activity directly from the X firehose.
• Academic Research: Using DeepSearch to cross-reference sources and synthesize complex technical information.

Development & Coding

• Complex Problem Solving: Using Reasoning Mode to tackle algorithmic challenges, math proofs, and system design questions.
• Code Review: Analyzing code for bugs, security vulnerabilities, and performance issues.
• API Integration: Building Grok-powered features directly into applications.

Content & Analysis

• Content Creation: Writing, editing, and brainstorming with unfiltered directness.
• Data Analysis: Interpreting charts, tables, and screenshots via vision capabilities.

Getting Started

Step 1: Access Grok

1. Visit grok.com or open the X (Twitter) mobile app.
2. Sign in with your X or xAI account.
3. Free users get access to standard Grok; SuperGrok unlocks higher usage limits and advanced features.

Step 2: Enable Reasoning Mode

1. Start a new conversation.
2. Click the "Think" toggle in the input bar to activate Reasoning Mode.
3. Submit your complex question — Grok will show its thinking process before answering.

Step 3: Use DeepSearch

1. Click the DeepSearch button (globe icon) to activate live web research.
2. Ask a question that requires current information (e.g., "What happened with the Fed rate decision today?").
3. Grok will cite its sources in the response.

Step 4: Access the API

1. Visit console.x.ai.
2. Create an API key.
3. Use the standard OpenAI SDK by pointing it to https://api.x.ai/v1.

Best Practices

• Use Reasoning Mode for math, logic, and multi-step coding tasks.
• Use DeepSearch when you need current news or verified facts.
• Be direct: Grok responds best to clear, specific prompts.

Pricing & Plans

• Free Tier: Daily usage limits, and not the flagship model. SpaceXAI does not publish which model the free tier serves.
• SuperGrok ($30/month): Longer conversations in Chat, higher usage on Grok Imagine, Voice Mode and Companions, access to the newest models, and early access to new features.
• SuperGrok Heavy: A higher tier with maximum usage limits and early feature access. Check grok.com for the current price.
• API (Pay-as-you-go) — prices in USD, from docs.x.ai/developers/pricing, checked 6 September 2026:
  • Grok 4.6: $2.00 per 1M input tokens, $0.50 cached, $6.00 per 1M output tokens — below 200K prompt tokens.
  • Grok 4.5: the same $2.00 / $6.00 headline, with a cheaper cached-input rate at $0.30.
  • Grok 4.3: $1.25 per 1M input tokens, $0.20 cached, $2.50 per 1M output tokens, at 1M context.
  • Grok Build 0.1 (coding): $1.00 per 1M input tokens, $2.00 per 1M output tokens.
  • Imagine API: from $0.02 per image (grok-imagine-image) to $0.04 (grok-imagine-image-2.0); video from $0.050 per second, or $0.080 per second on grok-imagine-video-1.5.
  • Voice API: speech-to-speech $0.08 per minute of audio; speech-to-text $0.10/hour REST or $0.20/hour streaming; text-to-speech $15.00 per 1M characters.

The long-context pricing cliff is the thing to plan around. Every text model steps to a higher rate once a prompt reaches 200,000 tokens, and the higher rate applies to every token in the request — input, cached and output alike — not just the tokens past the threshold. On Grok 4.6 that means $4.00 / $1.00 / $12.00. A 199K-token prompt and a 201K-token prompt are not 1% apart in cost; they are 100% apart. Either compact the context below the line or budget for double.

Priority Processing is a service tier, not a model. Requesting service_tier: "priority" applies a flat 2x multiplier on every token type of the same model ID, billed only when the response confirms the priority tier; caching discounts are applied before the multiplier. There is no grok-4.6-fast to call.

Limitations

• Primary Data Source: Real-time data is sourced primarily from X; broader web coverage is via DeepSearch only.
• No live knowledge by default: xAI notes that Grok has no knowledge of events beyond its training data unless you enable server-side search tools (Web Search / X Search).
• Availability: Some features require an X Premium or SuperGrok subscription, and the flagship model is not on the free tier.
• The 200K pricing cliff: crossing 200,000 prompt tokens doubles the rate on the whole request. It is invisible in the headline $2.00 / $6.00 figure and it is the dominant cost fact on long agent loops.
• Context shrank at the top: Grok 4.6 and 4.5 hold 500K tokens, against 1M on the older Grok 4.3. The newest model is not the one to reach for on a whole-corpus prompt.
• The consumer surface is undocumented: SpaceXAI publishes a precise API model table and says nothing about which model the Grok app or the X integration serves at each subscription tier.

Community & Support

• Official Website: grok.com
• API Docs: docs.x.ai
• X (Twitter): @xai for updates
• Discord: Official xAI Discord

Related Tools

• ChatGPT
• Claude
• Perplexity AI
• DeepSeek