Grok 4.7: price, context window and specs

Who makes Grok now

SpaceX announced on 2 February 2026 that it had acquired xAI, per the x.ai news post. The x.ai website and the API docs now present the company as SpaceXSI (the docs say "SpaceXSI (xAI)"), while the API itself is still called the xAI API and served at api.x.ai. We didn't find a separate rename announcement, so we use the name as the company's own pages show it.

What SpaceXSI says changed

Per the announcement, Grok 4.7 uses a new, larger base model than Grok 4.6 and a longer reinforcement learning run weighted toward tasks that take many hours. SpaceXSI says it's better at checking its own work and managing long context, and that it's served at the same price and speed as Grok 4.6. In the vendor's own table, Grok 4.7 at xhigh effort scores 46.3% on CursorBench 4.0, against 51.8% for Claude Fable 5.1 at max. Those are SpaceXSI's figures.

Pricing details

• Base: $2 per million input tokens, $0.50 cached input and $6 output while the prompt is under 200K tokens, per the models and pricing docs.
• Long context: once a prompt reaches 200K tokens, the whole request is billed at $4 / $1 / $12.
• Priority Processing: 2x standard rates, billed only when the response confirms the priority tier.
• US regional endpoint (us.api.x.ai): 1.1x, or $2.20 / $0.55 / $6.60 below 200K prompt tokens.
• The Batch API isn't supported for grok-4.7, so there's no batch discount.
• Server-side tools are billed on top of tokens, for example $5 per 1,000 web search calls.

Grok 4.7 Fast

Grok 4.7 Fast is the same model on faster infrastructure. The pricing docs list it at 2x the standard rates ($4 / $1 / $12 per million tokens), or 1.5x for long-context requests ($6 / $1.50 / $18). It's only in Cursor and Grok Build, billed through your plan there, isn't on the public xAI API and isn't included in Grok Build's free tier. The announcement describes it as twice the output speed at twice the price.

Things to know before calling the API

• On the Responses API, grok-4.7 always returns encrypted reasoning content. xAI's Grok 4.7 guide says to pass those reasoning items back unchanged in the next request.
• xAI strongly recommends setting a prompt_cache_key so a conversation's requests hit the same server; without it you often pay the full input price.
• logprobs and top_logprobs are ignored on grok-4.20 and newer models.
• The model has no knowledge of current events unless you enable the Web Search or X Search tools.

Independent benchmarks

• Artificial Analysis lists Grok 4.7 (xhigh) at 46 on its Intelligence Index v4.3.2, 30th of 216 models, at $3.74 per Intelligence Index task and about 86 output tokens per second, as of 13 October 2026. It notes the model is very verbose: 240M output tokens across the index.
• Vals AI lists Grok 4.7 at 54.95% (±1.07) on the Vals Index, 16th of 47 models, at $12.11 per test, as of 13 October 2026. Its launch-day note on 21 September reported 60.22% and 10th of 59 on the index at that time, tested at xhigh effort.