Grok developer guide (unofficial)

What is the Grok API?

Short answer: It's xAI's developer API for the Grok models, served at https://api.x.ai/v1. The current model is grok-4.7: text and image in, text out, with a 500,000-token context window and reasoning that is always on.

xAI documents two ways to send a request. The Responses API is the main one. Chat Completions is still available, but xAI calls it a legacy endpoint on its own page and labels it "Deprecated" in its API comparison table.

Is the Grok API OpenAI-compatible? Yes. xAI's docs use the OpenAI SDK with base_url="https://api.x.ai/v1" and your key from XAI_API_KEY. xAI also documents its own gRPC-based Python SDK, xai-sdk, and the AI SDK provider @ai-sdk/xai.

Grok 4.7 at a glance

Short answer: grok-4.7 costs $2 per million input tokens and $6 per million output tokens below 200K prompt tokens, has a 500,000-token context window, and starts at 150 requests per second on the lowest rate-limit tier.

Table: Grok 4.7 on the xAI API (last checked 8 Oct 2026)

What is Grok 4.7's knowledge cutoff? Sources differ. xAI's docs list a May 2026 knowledge cutoff; the Grok 4.7 model card gives a June 2026 pretraining-data cutoff. The two may measure different things, so we don't pick one.

Other models xAI still lists on the API include grok-4.6, grok-4.5, grok-4.3, grok-4.20-0309-reasoning, grok-4.20-0309-non-reasoning, grok-4.20-multi-agent-0309, grok-build-0.1, the Imagine image and video models, and the voice models. The full list, with context windows, is in Grok API models, pricing and rate limits.

Which Grok guide do you need?

Short answer: Start with the quickstart for your first call, the pricing page for costs and limits, and the channels page if you call Grok through a gateway. The tools pages cover function calling, JSON output and search.

• How do I make my first Grok API call? Grok API quickstart: your first grok-4.7 call with curl, Python and TypeScript
• Why am I getting a 401, 422 or 429 error? What Grok API error codes mean
• What does Grok 4.7 cost? Grok 4.7 pricing, including the 200K long-context rule
• What are my Grok API rate limits? xAI rate-limit tiers, RPS and TPM explained
• Why is my old model name now answered by another model? What happened to retired Grok models
• What is Grok 4.7's model ID on OpenRouter, Vercel or Cloudflare? Grok 4.7 model IDs and prices by channel
• Why does my tool call arrive in one chunk, or my stop parameter fail? Grok function calling and structured outputs: the quirks
• What replaced Live Search and search_parameters? Moving from Live Search to the web_search and x_search tools
• Using Grok as an app, not the API? See our Grok Bot guides for the desktop and mobile app.

What changed recently in the Grok API?

Short answer: Grok 4.7 launched on 21 Sep 2026, grok-voice-transcribe-1.0 reached end of life on 2 Oct 2026, a batch of older model names was retired on 15 May 2026, and grok-imagine-image-quality retires on 2 Nov 2026.

• 2 Nov 2026 (upcoming): grok-imagine-image-quality retires. After that, xAI serves those requests with grok-imagine-image-2.0 at quality: "low" and bills them at that rate.
• 2 Oct 2026: grok-voice-transcribe-1.0 reached end of life; requests to it go to grok-voice-transcribe-2.0 at the same price.
• 24 Sep 2026 (UTC): xai-sdk 1.20.0 published on PyPI and @ai-sdk/xai 5.0.8 published on npm.
• 21 Sep 2026: Grok 4.7 launched as grok-4.7. OpenRouter listed it as x-ai/grok-4.7 the same day.
• 18 Sep 2026: Grok Voice Transcribe 2.0 (grok-voice-transcribe-2.0) became available. It is now the default, and grok-voice-transcribe-1.0 reached end of life on 2 Oct 2026.
• 2 Sep 2026: xAI began its 60-day retirement notice for grok-imagine-image-quality.
• 15 May 2026, 12:00 PM PT: xAI retired older slugs, including grok-4-fast-reasoning, grok-4-1-fast-reasoning, grok-4-0709, grok-code-fast-1 and grok-3. Requests to them are now redirected to newer models.
• As of 8 Oct 2026: xAI's old Live Search guide URL redirects to its Web Search tool page.

Is Grok the same thing as Grok Bot?

Short answer: No. The Grok API is for developers who send requests from code. Grok Bot is a separate desktop and mobile app, and it has its own Grok Bot guides.

This hub covers the API only: model IDs, prices, limits, SDKs and tools.

Related guides across tools

Short answer: For the same model on other platforms, and for prices across vendors, use the comparison pages.

• Jev model IDs by channel, the same channel-by-channel format for Jev

FAQ

Is the Grok API OpenAI-compatible?

Yes. xAI's docs use the OpenAI SDK with base_url="https://api.x.ai/v1" and your key from XAI_API_KEY. xAI also documents its own gRPC-based Python SDK, xai-sdk, and the AI SDK provider @ai-sdk/xai.

What is the Grok API?

It's xAI's developer API for the Grok models, served at https://api.x.ai/v1. The current model is grok-4.7: text and image in, text out, with a 500,000-token context window and reasoning that is always on.

How much does Grok 4.7 cost?

grok-4.7 costs $2 per million input tokens and $6 per million output tokens below 200K prompt tokens, has a 500,000-token context window, and starts at 150 requests per second on the lowest rate-limit tier.

What is Grok 4.7's knowledge cutoff?

Sources differ. xAI's docs list a May 2026 knowledge cutoff; the Grok 4.7 model card gives a June 2026 pretraining-data cutoff. The two may measure different things, so we don't pick one.

Is Grok the same thing as Grok Bot?

No. The Grok API is for developers who send requests from code. Grok Bot is a separate desktop and mobile app, and it has its own Grok Bot guides.