Models

Grok 4.7: xAI’s Model for Current Information, Coding, and Lower-Cost Tasks

Grok is the model family from xAI, Elon Musk’s company. Its strengths are access to the stream of posts on X (formerly Twitter) and aggressive pricing for its capabilities. In fall 2026, Grok 4.7 is the flagship for coding, agent tasks, and knowledge work, with a 500,000-token context. Here is how it differs from GPT and Claude, and where its limits are.

Current Grok models

Grok 4.7 launched on September 21, 2026, on a larger base model. It had a longer reinforcement-learning cycle on a more demanding task mix, which is intended to improve self-checking and long-context work. xAI describes it as “twice as fast and half the price” of comparable models. By token price, it is one of the better-value flagship offers so far.

Its context window is 500,000 tokens. It accepts text and images and has configurable effort levels (low, medium, high, and xhigh). The model is available not only from xAI but also on Amazon Bedrock and Google’s platform.

ModelRolePrice per 1M tokens
Grok 4.7Flagship for coding, agent tasks, and knowledge work$2 input / $6 output (same price as 4.6)
Grok 4.7 FastFast version for code editorsLower cost; available in Cursor and Grok Build
Grok 4.6Previous flagship$2 input / $0.50 cached input / $6 output

How capable is it?

Benchmarks show Grok 4.7 improving on 4.6 across the main metrics and performing comparably to other frontier models. It competes with GPT and Claude without making sweeping claims of superiority. Its strengths are long tasks that require the model to check its own work, and coding.

xAI also offers voice models. Grok Voice Transcribe 2.0, released in September 2026, was described as one of the most accurate transcription systems, with automatic language detection even for short phrases. Audio transcription is currently one of xAI’s stronger specialties.

Safety and refusals

Grok has historically been positioned as a less restrictive model with access to the live X feed. In its Grok 4.20 system card, xAI reported reducing sycophancy and making the model more likely to correct users who insist on an incorrect view. That is a positive sign.

Access to X data also creates a problem: answer quality depends on a social feed full of noise and misinformation. Grok may confidently cite recent posts, including dubious ones. Verify factual claims. Grok usually refuses gray-area requests less often than Claude, while the X chat and API may follow different rules; read the terms of the service you use.

What does a task cost?

$2 per million input tokens and $6 per million output tokens, with cached input at $0.50, is notably cheaper than GPT Sol ($5/$30) and Claude Opus 5.5 ($4/$20). If a task does not need the very highest capability, Grok may offer a better balance of price and quality.

Cost per task is harder to estimate because it depends on how much reasoning is involved. Grok 4.7 checks its work more actively than its predecessor and may use more reasoning tokens on difficult tasks. Track actual usage rather than relying only on the rate card.

How Grok works

  • A 500K-token context is smaller than Gemini’s or Claude’s, but enough for most documents.
  • Configurable reasoning effort: low, medium, high, or xhigh.
  • Input: text and images. Output: text only.
  • X integration can bring current discussions and news into answers when search is enabled.
  • Voice tools include transcription and an assistant, including in Tesla vehicles.
  • Limit: its tools and integrations ecosystem is smaller than OpenAI’s or Google’s.

Who should use Grok?

Grok is a sensible option when price and current information matter: news monitoring, discussion analysis, or bulk text processing. For tightly regulated work such as legal or medical documents, people often choose Claude or GPT for more predictable behavior.

Grok appears in aggregator catalogs from time to time, including ours; check the current list before you start. To compare it with Western flagships on the same prompts, begin with our GPT model guide, which explains when Sol may justify its higher price.

Frequently asked questions

Is Grok available only on X?

No. There is a standalone Grok app, the xAI API, Amazon Bedrock, and other platforms. Terms and limits vary.

Why is Grok cheaper than its competitors?

xAI uses aggressive pricing to gain market share. Grok 4.6 and 4.7 cost less per token than flagship GPT and Claude models.

Is Grok less censored?

It has historically refused less often, but it still follows xAI rules and its Acceptable Use Policy. A “less restrictive” label does not remove model errors or legal limits.

Does Grok know current news?

Search and access to X can bring recent discussions into its answers. Source quality on social media varies, so verify factual claims.

Grok or Gemini for large documents?

Gemini has a context of up to 10 million tokens; Grok has 500K. For very large inputs, the difference is clear. Both can handle ordinary documents.

Sources

How we check our guides