Grok 4: Pricing, Benchmarks & Access Guide [2026]

Grok 4 launched in July 2025 as xAI’s flagship reasoning model, with native tool use, real-time search, and access through both Grok and the xAI API.

The Grok 4 family has expanded since then. Grok 4.5 launched in July 2026 and is now the newer Grok frontier model for coding, agentic tasks, and knowledge work.

This guide covers Grok 4 pricing, plans, API access, benchmarks, context limits, Grok 4 Heavy, and newer Grok 4-family models so you can see what is available in 2026 and which option fits your use case.

What Is Grok 4?

Grok 4 is a multimodal reasoning model released by xAI in July 2025. It accepts text and image inputs, produces text output, and supports a 256,000-token context window through the API.

Compared with Grok 3, Grok 4 put more emphasis on reasoning and native tool use:

• Context window: 256,000 tokens
• Inputs: Text and image
• Output: Text
• Reasoning: Trained with reinforcement learning for complex reasoning tasks
• Native tools: Web search, X search, and code execution

xAI trained Grok 4 to use tools while reasoning. For tasks that need current information or additional computation, the model can use web search, search X, or work with code rather than relying only on its training data.

xAI later published a Grok 4 model card covering the model’s safety evaluations, capabilities, and limitations.

Grok 4 Benchmarks

At launch in July 2025, Artificial Analysis reported a score of 73 for Grok 4 on its Intelligence Index at the time, ahead of OpenAI o3 and Gemini 2.5 Pro at 70, DeepSeek R1 0528 at 68, and Claude 4 Opus at 64.

Those results are best treated as historical launch benchmarks because Artificial Analysis has since updated its benchmark suite and scoring methodology.

In xAI’s Grok 4 launch testing, Grok 4 reached 15.9% on ARC-AGI-2, while Grok 4 Heavy reached 50.7% on Humanity’s Last Exam and 61.9% on USAMO 2025.

How to Access Grok 4

Access to Grok 4 has changed since its July 2025 launch. Grok is now available through Grok.com, X, iOS, and Android, while the xAI API lineup has moved to newer Grok 4-family models.

1. Grok.com and SuperGrok Plans

Grok now has a free plan, while SuperGrok costs $30 per month and provides higher usage limits plus access to frontier models such as Grok 4.5. xAI also lists SuperGrok Lite, SuperGrok Plus, and SuperGrok Heavy among its current plan options.

Grok 4 Pricing: How Much Does It Cost?

Grok pricing now depends on whether you use the consumer Grok app or the xAI API.

For individual users, xAI’s current Grok pricing includes:

• Free: $0/month with limited access to Grok features.
• SuperGrok: $30/month with higher usage limits and access to Grok 4.5.
• SuperGrok Plus: $100/month with significantly higher usage, priority access, and additional features.

xAI also lists SuperGrok Lite and SuperGrok Heavy among its plan options, although the public pricing page does not currently display a price for those tiers.

API pricing is separate from Grok subscriptions.

Under xAI’s current API pricing, Grok 4.3 costs $1.25 per million input tokens and $2.50 per million output tokens for short-context requests. Grok 4.5 costs $2 per million input tokens and $6 per million output tokens, with higher rates when the prompt reaches its long-context pricing threshold.

2. Standalone Web Interface

Website: Grok.com is the official web interface for Grok. According to xAI’s current Grok documentation, Grok is available on the web at grok.com and through its iOS and Android apps.

When you sign in with the same account, your conversations, settings, and subscription stay synchronized across those platforms. Available models, features, and usage limits depend on your current Grok plan.

3. Mobile Applications

Platforms: Grok is available through official iOS and Android apps. The models and usage limits available in the apps depend on your Grok plan.

4. API Access

xAI provides Grok models through its API for developers building applications, agents, and other AI workflows. The current API lineup includes newer Grok 4-family models such as Grok 4.3, Grok 4.20, and Grok 4.5.

The original grok-4-0709 API model was retired on May 15, 2026. xAI now redirects requests using that model slug to Grok 4.3 and recommends Grok 4.3 for workloads that previously used the original Grok 4 reasoning model. xAI explains the migration here.

Grok 4.3 supports a 1 million-token context window and currently costs $1.25 per million input tokens and $2.50 per million output tokens. Grok 4.5 supports a 500,000-token context window and starts at $2 per million input tokens and $6 per million output tokens. xAI applies higher rates when requests cross each model’s long-context pricing threshold.

Is Grok 4 Free?

Grok is free to start, but that does not mean every Grok model is available on the free plan. xAI’s current pricing page lists a $0 Free plan with limited access to Grok features, while Grok 4.5 is specifically listed with the $30/month SuperGrok plan.

xAI does not currently list the original Grok 4 model as a specific Free-plan entitlement. So if you are searching for “Grok 4 free,” the safer distinction is that Grok has a free tier, while access to particular frontier models depends on the plan and current model lineup. xAI’s documentation likewise states that Grok is free to start and that paid SuperGrok plans raise limits and unlock more capabilities.

For developers, API access is usage-based rather than part of the consumer free plan.

Grok 4 Context Window and Token Limits

The original Grok 4 API launched with a 256,000-token context window, according to xAI’s Grok 4 launch announcement.

Newer Grok 4-family API models support different limits:

• Grok 4.3: 1,000,000 tokens
• Grok 4.20: 1,000,000 tokens
• Grok 4.5: 500,000 tokens

So there is no single “Grok token limit” that applies to every model. The context window depends on the specific Grok model you use. A larger context window lets a model process more text, code, conversation history, or document content in one request, but it does not by itself guarantee better output quality.

When Was Grok 4 Released?

Grok 4 was released on July 9, 2025. xAI announced Grok 4 that day alongside Grok 4 Heavy, with access through SuperGrok, X Premium+, and the xAI API at launch.

The Grok lineup has changed since then, but July 9, 2025 remains the original Grok 4 release date.

A Timeline to Grok 4: From Grok-1 to Now

xAI has moved through several major Grok generations since Grok-1 launched in 2023. Grok 4 arrived in July 2025, followed by newer Grok 4-family models that expanded the lineup through 2026.

• November 2023: Grok-1 launches. xAI released its first Grok model on X, with a focus on question answering, information retrieval, writing, and coding assistance.
• March 2024: Grok-1 goes open source. xAI released the Grok-1 base model weights and architecture under the Apache 2.0 license.
• August 2024: Grok-2 arrives. Grok-2 improved chat, coding, reasoning, and vision capabilities and began rolling out on X.
• November 2024: xAI opens its API public beta. Developers gained access to Grok models through xAI’s API platform.
• February 2025: Grok 3 launches. xAI introduced Grok 3 with stronger reasoning, math, coding, and instruction-following capabilities after training it on the Colossus supercluster.
• July 2025: Grok 4 launches. Grok 4 introduced native tool use and real-time search integration, alongside the more compute-intensive Grok 4 Heavy option.
• November 2025: Grok 4.1 launches. xAI released Grok 4.1 with improvements aimed at real-world conversations, instruction following, and interaction quality.
• May 2026: the original Grok 4 API model is retired. xAI retired grok-4-0709, with requests to that model redirected to Grok 4.3.
• July 2026: Grok 4.5 launches. xAI introduced Grok 4.5 as its newer frontier model for coding, agentic tasks, and knowledge work.

What Is Grok 4.20?

Grok 4.20 is a newer Grok 4-family API model. It is not another name for Grok 4 Heavy.

xAI describes Grok 4.20 as a high-performance model built for agentic tool calling and strong prompt adherence. It accepts text and image input, produces text output, and supports a 1 million-token context window.

Grok 4.20 is available in reasoning and non-reasoning variants. xAI also offers a multi-agent version in which multiple agents can work in parallel on research tasks.

For standard-context API requests, xAI currently prices Grok 4.20 at $1.25 per million input tokens and $2.50 per million output tokens. Long-context requests use higher rates.

What Is Grok 4 Heavy?

Grok 4 Heavy is the higher-compute version of the original Grok 4 that xAI introduced in July 2025. Unlike Grok 4.20, it was designed around parallel test-time compute, allowing the system to consider multiple approaches to a problem before producing an answer.

In xAI’s Grok 4 launch announcement, Grok 4 Heavy reached 50.7% on Humanity’s Last Exam and 61.9% on USAMO 2025.

xAI’s current Grok pricing page still lists SuperGrok Heavy among its subscription options.

Why Choose Grok 4?

Grok 4 was a major step forward in xAI’s model family, particularly for multimodal reasoning and tool-assisted tasks. Its main strengths at launch included:

• Large context window: The original Grok 4 API supported 256,000 tokens, giving it room to work with long documents, code, and conversation history.
• Text and image input: Grok 4 could reason across both text and visual inputs rather than operating as a text-only model.
• Native tool use: xAI built Grok 4 to work with tools such as web search, X search, and code execution while reasoning.
• Access to current information: xAI integrated real-time search across X, the web, and news sources into the Grok 4 API.

Grok 4 vs. Grok 3: What Changed?

If you are currently using Grok 3, Grok 4 changed several parts of xAI’s model approach:

• Context window: xAI launched Grok 3 with a 1 million-token context window. The original Grok 4 API launched with a 256,000-token context window, so Grok 3 had the larger raw context limit at launch.
• Multimodal understanding: Grok 3 already demonstrated image and video understanding. Grok 4 continued multimodal support across text and vision rather than introducing vision for the first time.
• Reasoning: Grok 3 offered a dedicated Think mode that users could activate for harder problems. Grok 4 was trained with reinforcement learning at a larger scale to strengthen reasoning across more domains.
• Tool use: Grok 3 introduced capabilities such as DeepSearch and access to code interpreters and the internet. Grok 4 pushed this further with native tool use and real-time search across the web, X, and news sources.
• Benchmarks: xAI reported strong results for both generations, but the benchmark sets and evaluation conditions differ. Grok 4’s launch results included 15.9% on ARC-AGI-2, while Grok 3’s launch testing focused on benchmarks such as AIME, GPQA, LiveCodeBench, and MMMU.

Bottom line: Grok 4 was not simply a larger-context replacement for Grok 3. Grok 3 actually launched with the bigger context window, while Grok 4 put greater emphasis on advanced reasoning, multimodal work, native tools, and real-time search.

What Is Grok 4.5?

Grok 4.5 launched on July 16, 2026 as xAI’s newer model for coding, agentic tasks, and knowledge work. xAI’s Grok 4.5 announcement describes it as its smartest model at launch for these workloads.

Compared with the original Grok 4, Grok 4.5 is aimed more directly at software engineering and multi-step agentic work. Its current API specifications include:

• Context window: 500,000 tokens
• Inputs: Text and images
• Output: Text
• Reasoning: Supported
• API price: $2 per million input tokens and $6 per million output tokens

You can see the current specifications on the official Grok 4.5 model page. xAI applies higher input rates to requests that exceed the model’s 200,000-token long-context pricing threshold.

Grok 4.5 is also available through Grok.com, X, iOS, and Android, so it is not limited to API users. xAI expanded Grok 4.5 across those consumer platforms on July 22, 2026.

Grok 4 Models Compared

For new API projects, xAI currently recommends Grok 4.5 for code and general-purpose work. Grok 4.20 remains available for workloads that benefit from a 1 million-token context window and agentic tool calling.

Grok 4 for Coding

Grok 4 was built for complex reasoning and supported coding tasks through native tool use, including code execution. At launch, xAI also evaluated Grok 4 on LiveCodeBench, a competitive coding benchmark.

For new coding projects in 2026, though, xAI now recommends Grok 4.5 rather than the original Grok 4. Its current model guidance names Grok 4.5 as the preferred model for code and general-purpose work.

Grok 4.5 was trained specifically for coding, software engineering, agentic tasks, science, engineering, and math. In xAI’s launch testing, it scored 64.7% on SWE-Bench Pro, 83.3% on Terminal Bench 2.1, and 29.0% on SWE Marathon. xAI’s Grok 4.5 announcement also shows examples of the model building end-to-end applications from a single prompt.

So if you are comparing Grok models mainly for coding, the original Grok 4 remains an important part of the model family’s development, but Grok 4.5 is the current xAI choice for new coding and agentic software workflows.

Using Grok With Chatbase

Choosing a Grok model determines the reasoning engine behind an AI agent. But the model alone does not provide your business knowledge, workflows, deployment channels, or human support process.

Chatbase currently lists Grok 4 as an available model for AI agents, so you can use Grok as the reasoning model and connect it to the rest of your agent setup.

Chatbase acts as the layer around the model. Once you choose a supported model, you can connect the agent to the parts it needs to work in a real customer-facing environment:

• Business knowledge: Add websites, documents, text, Q&As, Notion pages, and supported ticket sources through Chatbase data sources.
• Actions: Let the agent trigger tasks, retrieve information, or hand a conversation to a person using AI Actions.
• Procedures: Define step-by-step workflows that the agent follows when a specific customer request matches a procedure. Chatbase Procedures can also control when particular actions are used.
• Channels: Deploy the same agent across supported channels such as your website, email, WhatsApp, Messenger, Slack, and phone through Chatbase Channels.
• Human support: When a conversation needs a person, Chatbase can hand it off or manage the resulting ticket through its Helpdesk.

This is the practical difference between choosing an LLM and deploying an AI agent. Grok provides the underlying model capabilities. Chatbase connects the model to the knowledge, tools, workflows, channels, and support team around it.

Ready to Get Started with Grok?

Grok 4 remains an important part of xAI’s model family, but the lineup has changed significantly since its 2025 launch. Grok 4.5 is now xAI’s recommended model for coding and general-purpose API work, while Grok 4.3 and Grok 4.20 serve other workloads across the current Grok lineup.

Which model makes sense depends on the workload. Grok 4.3 and Grok 4.20 offer 1 million-token context windows, while Grok 4.5 focuses more heavily on coding, agentic software tasks, engineering, and knowledge work with a 500,000-token context window.

If you want to use Grok inside a customer-facing AI agent, Chatbase currently lists Grok 4 as a supported model and lets you connect the agent to your own business data and surrounding workflows.

Build an AI agent with Grok 4