Grok 4 Review: Specs, Pricing & Key Features

Grok 4 is the large language model xAI released on July 9, 2025, and at launch Elon Musk called it the most intelligent model in the world. It brought frontier-level reasoning, native tool use with real-time search across the web and X, and Grok 4 Heavy, a multi-agent version built for the hardest problems.

This Grok 4 review covers what the model is, what its benchmarks showed, how it compares with later Grok releases, what it cost, and what changed after xAI retired it. If you want to use xAI's models today, Grok 4.5 is available in Chat Smith on Android, iOS, and the web.

Grok 4 is the fourth-generation model from xAI, Elon Musk's AI company. xAI skipped the expected Grok 3.5 and went straight to Grok 4, training it on Colossus, its supercomputer cluster of about 200,000 GPUs. The big change was scale: xAI applied reinforcement learning at pretraining scale so the model would reason through problems rather than simply predict the next word.

Grok 4 is a reasoning-only model, so it always thinks before it answers. It was also trained to use tools on its own, including a code interpreter and web browsing, and it can search X for real-time posts and news. Through the API, it offered a 256,000-token context window.

Grok 4 is no longer xAI's current model. The original grok-4-0709 API model was retired on May 15, 2026, and xAI now routes requests to that legacy identifier to Grok 4.3. For where it sits in the family, see our full list of Grok models or the Grok 3 vs Grok 4 comparison.

Four capabilities defined Grok 4 at launch.

At launch, Grok 4 posted some of the strongest reasoning scores of any model. It reached 87.5% on GPQA Diamond, a graduate-level science test, and 15.9% on ARC-AGI-2, nearly double the previous best among closed models. With tools, Grok 4 Heavy became the first model to pass 50% on the text-only subset of Humanity's Last Exam, scoring 50.7%, and it led the USAMO 2025 math olympiad test at 61.9%. That made it a strong fit for multi-step math and science, the kind of work an AI math solver breaks down step by step.

Instead of relying on plugins, Grok 4 was trained with reinforcement learning to decide when to search, browse, or run code. It can pull live information from the web and from X, then synthesize it into one answer, which made it unusually good at fast-moving topics like news, markets, and online trends. It is a more autonomous version of what an AI article summarizer does with a single link.

Grok 4 Heavy runs several agents in parallel on the same problem, then compares their work and combines the best answer. This test-time compute approach costs more but reduces errors on the hardest tasks. It was reserved for the SuperGrok Heavy plan at $300 per month.

Grok 4 accepts both text and images, so it can read screenshots, charts, and photos. In the Grok app, xAI paired it with an upgraded voice mode that lets users point their phone camera at a scene and talk about what Grok sees.

Grok 4 opened the Grok 4 line, and xAI has iterated quickly since. The table compares it with Grok 4.5, the version available in Chat Smith, and Grok 4.7, xAI's newest flagship as of September 2026.

Both successors are cheaper per token and offer larger context windows than Grok 4. Our Grok 4.5 review and Grok 4.7 review cover them in detail.

Grok 4 launched in the xAI API at $3 per million input tokens and $15 per million output tokens, with rates doubling for prompts over 128,000 tokens. For consumers, it was part of xAI's paid plans:

Today, the original model is gone from the API, and the lower-cost Grok 4 Fast line and newer flagships have taken its place. In Chat Smith, you can use Grok 4.5 without an xAI account: free accounts include limited use, and Chat Smith Pro raises usage limits for $9.99 per month or $39.99 per year.

The main issue is simple: Grok 4 has been retired. Anyone calling the old grok-4-0709 identifier now gets Grok 4.3 at low reasoning effort, so results labeled "Grok 4" may not come from the original model.

Even at launch, Grok 4 had trade-offs. As a reasoning-only model, it could be slow on simple questions. Its training data stopped in November 2024, so it relied on live search for anything newer, and several headline scores came from Grok 4 Heavy with tools rather than the standard model.

Its 256,000-token context window and $3/$15 pricing also look dated next to its successors, which offer larger windows at lower prices. If you are choosing a model today, compare current Grok releases with rivals on your own tasks rather than relying on Grok 4's 2025 benchmarks.

Chat Smith brings xAI's current models into one app on Android, iOS, and the web, with no xAI account or Grok subscription needed. Pick Grok 4.5 from the model list and start chatting.

Grok 4.5 carries forward Grok 4's strengths in reasoning and research at a lower cost per token, with a larger 500,000-token context window. You can also put the same question to Claude Sonnet 5 or DeepSeek V4 Pro and keep the answer that holds up.

For questions that need up-to-date sources, Chat Smith's built-in Deep Research feature runs advanced search and reasoning across the web, a similar idea to the real-time research that made Grok 4 stand out.

Grok 4 was a milestone for xAI: frontier reasoning scores, native tool use with real-time search, and a multi-agent Heavy mode that pointed the way for later releases. As a product, though, it has run its course, and xAI has retired it in favor of faster, cheaper, and stronger models.

If you want the Grok experience today, start with Grok 4.5. You can try it now in Chat Smith, an AI chat app that puts it next to the latest models from OpenAI, Anthropic, Google, and DeepSeek.