Drafted with AI tools and edited by a human. The figures and conclusions were checked by a GPUniq editor.
TL;DR
Grok 4.7 supports a 500,000-token context window. Pricing matches Grok 4.6 for inputs under 200,000 tokens: $2 per million input tokens and $6 per million output tokens. Over 200,000 tokens, rates double to $4 per million input tokens and $12 per million output tokens. Cached tokens are cheaper. The Fast variant, available only in Cursor and Grok Build, is billed at about twice the standard rates. Public benchmarks: 62.4 percent on LatchBio biosafety, about 3.3 percent risky prompt pass rate on HackerBench v0.3.
Grok 4.7 pricing
Grok 4.7 uses these rates:
| Variant | Input Price ($/1M) | Output Price ($/1M) | Context Limit (tokens) |
|---|---|---|---|
| Standard (<200k) | 2 | 6 | 500,000 |
| Standard (>=200k) | 4 | 12 | 500,000 |
| Fast (<200k) | 4 | 12 | 500,000 |
| Fast (>=200k) | 8 | 24 | 500,000 |
Cached input tokens are $0.5 per million under 200k, $1 per million over 200k.
The Fast variant, only in Cursor and Grok Build, is priced at roughly double the standard rates.
Long context handling
Grok 4.7 allows up to 500,000 tokens per context. The cost doubles once you exceed 200,000 input tokens. For code analysis, multi-document summarization, or long conversations, it works, but you pay more as you increase context size.
Vendor documentation says there is "no limit" to text output, but latency and budget are practical constraints.
Benchmarks
Public results:
| Model | LatchBio (%) | HackerBench (%) | Context (tokens) |
|---|---|---|---|
| Grok 4.7 | 62.4 | 3.3 | 500,000 |
Grok 4.7 scores ahead of Grok 4.6, Fable 5.1, and GPT-6 Astra on GDPval, AA Briefcase, and EEBench, according to xAI. Only the LatchBio and HackerBench numbers are published.
Input and output modalities
Inputs: text and images. Outputs: text only. No image, audio, or video generation. Structured output (like JSON) is supported. Function and tool calling is available.
No batch API on public endpoints.
Developer access
You can access Grok 4.7 via:
- xAI API (model id: grok-4.7)
- Cursor and Grok Build (Standard and Fast variants)
- Third-party platforms: Cloudflare, OpenRouter, Vercel
- US regional endpoints
Limitations
- Costs jump sharply above 200,000 input tokens.
- Outputs are text only. Images accepted as input, not output.
- No batch API on the standard public API.
- About 3.3 percent pass rate on risky prompts (HackerBench v0.3).
- Fast variant is billed at about double the standard rate.
- Knowledge cutoff: May 2026.
API configuration
API lets you set reasoning effort: low, medium, high (default), or xhigh. You can select Standard or Fast variant. Context window is flexible up to 500,000 tokens, but pricing changes above 200,000. Structured outputs and tool-calling are supported. Watch your token usage to avoid higher charges.
Related on the GPUniq blog
Methodology
Facts are compiled from the vendor's own announcement and documentation, linked below, at the time of writing; where the vendor hasn't published a number it is marked as such. GPUniq prices, when the model is already available here, come from our live catalog.
Sources
- 1.Grok Models & Pricing | SpaceXAI Docs — SpaceXAI Docs (accessed )
- 2.Release Notes | SpaceXAI — SpaceXAI Docs (accessed )
- 3.Introducing Grok 4.7 — xAI / SpaceXAI News (accessed )
- 4.AWS Bedrock Model Card for Grok 4.7 — Amazon AWS Docs (accessed )
- 5.Grok 4.7 pricing: The $2/$6 card and the real bill — Tabbit Browser / Blog (accessed )
Want cheap GPUs for your next project?
Browse live GPU prices and rent the right card in seconds — H100, A100, RTX 4090, and 50+ more models.
