Grok Pricing in July 2026:
SuperGrok, Heavy, Free Tier
and API Costs

Grok costs from $0 to $300 per month. The free tier runs on grok.com and inside X with limited access. Paid consumer plans are X Premium at $8, SuperGrok Lite at $10, SuperGrok at $30, X Premium+ at $40, and SuperGrok Heavy at $300. API pricing for the current grok-4.3 flagship is $1.25 per million input tokens and $2.50 per million output.

Full breakdown of each tier, the tier-to-model transparency gap, and the complete API rate table below.

Grok’s free tier is accessible through grok.com and X without a paid subscription. The headline limit reported across independent sources is approximately 10 prompts every 2 hours. In practice, that translates to roughly 120 prompts per day if you used every refresh window, but most users hit the limit in clusters and wait for the timer rather than spreading queries evenly.

The free tier is best read as a sampling tier. It demonstrates Grok’s interaction style and surfaces the company’s distinctive features (real-time X access, the conversational personality), but the rate limits make it impractical for sustained workflows.

Grok access comes through two parallel subscription paths: standalone Grok (SuperGrok at $30/month, SuperGrok Lite at $10/month, SuperGrok Heavy at $300/month) and X-platform-bundled (X Premium at $8/month, X Premium+ at $40/month). The X-bundled tiers grant Grok access as a secondary benefit of an X subscription whose primary value is X platform features. The standalone tiers focus on Grok itself.

The X Premium and X Premium+ tiers are best understood as “Grok plus X stuff” rather than “Grok subscription.” If you do not value the X platform features, the standalone SuperGrok tiers are cheaper for the same Grok access.

The tenfold price gap between SuperGrok and SuperGrok Heavy reflects compute cost rather than feature breadth. The features that justify Heavy are concentrated in three areas:

This is the documented opacity in Grok’s pricing structure, and the question almost no published comparison answers. SuperGrok at $30/month is described as “Grok 4.3 rolling out in stages.” Tier-equivalent users receive different model variants at the same time. No UI indicator confirms which model processed any given query.

The mechanism behind the opacity

• Auto Mode routes dynamically. When a user submits a query, Grok’s Auto Mode selects an underlying model variant based on perceived query complexity. The selection is not exposed in the consumer UI.
• Staged rollouts split tiers. A new model release reaches different users at different times. Two SuperGrok subscribers can submit identical queries and hit different model versions during rollout.
• Mode aliases mask versions. The API supports model aliases. A retired slug like grok-4 now silently redirects to grok-4.3 and bills at grok-4.3 rates. For users calling aliases, migration to newer checkpoints happens without notification.

The only firm disambiguation path is API use with dated model IDs (e.g., grok-4.20-multi-agent-0309). Consumer app users cannot reliably determine which variant their query hit. If your workflow depends on knowing the model version, the API is the answer.

What this means in practice for each tier

For firm model disambiguation, use the API with dated model IDs (e.g., grok-4.20-multi-agent-0309). The consumer apps do not expose which model variant served any given query. No official xAI statement addresses this transparency gap directly.

The API exposes five active models with distinct input, cached input, and output rates. Pricing is per million tokens. Cached input applies to repeated context that has been previously processed.

grok-build-0.1 is a dedicated coding model (the engine behind Grok Build), priced below the general flagship at $1.00 / $2.00 per 1M with a 256K context window. The three grok-4.20 variants share grok-4.3 pricing at 1M context. Rates shift with little notice, so verify at console.x.ai before you rely on any figure for budgeting.

Free credits: New API accounts receive $25 in standard trial credits. A separate $150/month data-sharing promotion once pushed the historical maximum to $175, but that program should be treated as ended for new accounts. Promotional terms change often, so verify at console.x.ai.

API Tools and Storage

Batch API: Asynchronous processing within a 24-hour window receives a 20-50% discount on standard rates. Storage charges began in April 2026. Text-to-Speech was raised from its $4.20 launch price to the current $15.00 per 1M characters during the Q2 2026 API restructuring.

xAI’s official documentation states that “model access might vary depending on various factors such as geographical location, account limitations, etc.” Specific blocked countries are not enumerated. The documented restrictions:

• Memory feature was not available in certain unspecified regions at the April 2025 beta launch.
• NSFW Companions are likely geo-restricted in some jurisdictions; specific countries are not disclosed.
• Mainland China access requires VPN or proxy services per developer community reports.
• Sanctioned jurisdictions (Russia, Iran, DPRK, etc.) follow standard US export controls. No explicit xAI statement on jurisdiction-level restrictions was found.
• EU and UK have no documented GDPR-specific restrictions. xAI holds SOC 2 Type 2, GDPR, and CCPA certifications per the official Grok 4 announcement.
• Microsoft Azure AI Foundry offers Grok models for enterprise deployment (initially Grok-3 and Grok-3 Mini, since succeeded by current versions as those models were deprecated), providing an alternate access path for organizations with existing Azure relationships.

For users in regions where direct grok.com access is restricted, Azure AI Foundry is the documented enterprise channel. For developers, the API is accessible from most jurisdictions where Azure is available, with the caveat that specific country-level enforcement may vary.

The pattern is downward pricing pressure on flagship reasoning models with each generation, paired with aggressive consolidation. The May 15 retirement collapsed eight separate endpoints into grok-4.3, so most legacy slugs now bill at the flagship rate rather than their old, lower prices.