Grok 4.7 pricing is $2.00/1M input and $6.00/1M output per 1M tokens when the prompt stays under 200K. Cached input is $0.500/1M. At or above 200K prompt tokens, the whole request bills at the long-context band. Rates in TokenCalculator checked 23 Sep 2026.
Price a Grok workload
Grok counts are Approx in-browser. Use the cost calculator for planning, then confirm critical budgets against xAI usage meters.
Grok 4.7 rate card
API model id: grok-4.7. Context window: 500K tokens. Knowledge cut-off May 2026 per catalog notes. Prefer Grok 4.7 over Grok 4.6 for new budgets.
| Line | USD per 1M tokens |
|---|---|
| Input (<200K prompt) | $2.00/1M |
| Cached input (<200K) | $0.500/1M |
| Output (<200K) | $6.00/1M |
| Long-context input (>=200K, whole request) | $4.00/1M |
| Long-context cached input (>=200K) | $1.00/1M |
| Long-context output (>=200K, whole request) | $12.00/1M |
The 200K long-context cliff
When prompt tokens hit 200K or more, xAI reprices the entire request at $4 / $1 cached / $12 output per 1M. That is earlier than OpenAI GPT-6 Sol 272K cliff. Design agents around 200K before you treat the full 500K window as a flat price.
How Grok 4.7 compares nearby
| Model | Input / 1M | Output / 1M | Cache read / 1M | Cliff |
|---|---|---|---|---|
| Grok 4.7 (<200K) | $2.00/1M | $6.00/1M | $0.500/1M | 200K |
| Grok 4.6 (<200K) | $2.00/1M | $6.00/1M | $0.500/1M | 200K |
| GPT-6 Sol (<=272K) | $2.00/1M | $10.00/1M | $0.200/1M | 272K |
| Claude Opus 5.5 | $4.00/1M | $20.00/1M | $0.200/1M | none on card |
Under 200K, Grok matches Sol on input ($2) and undercuts Sol on output ($6 vs $10). Cache reads are higher on Grok ($0.50 vs $0.20). Cost per finished task can still flip if Grok burns more tokens.
Worked examples
- Short coding turn: 4K input + 1K output about $0.014 under 200K.
- Cached agent turn: 40K cached + 2K fresh + 1K output about $0.020 + $0.004 + $0.006 = $0.030.
- Prompt at 200K+: use the long-context band for the whole request.
When to use Grok 4.7
Use Grok 4.7 when xAI quality wins your harness for coding or long-running agents and you accept Approx token planning. Prefer Sol when Exact OpenAI counts and cheaper cache reads matter. Prefer Opus 5.5 when Claude tooling wins.
Common mistakes
- Shopping only on $2 / $6 and ignoring the 200K cliff
- Assuming cheaper output stickers mean cheaper agent bills
- Budgeting Grok with Exact OpenAI token counts
- Leaving new spend on Grok 4.6 when 4.7 is the preferred flagship row
Frequently asked questions
How much does Grok 4.7 cost?
Under 200K prompt tokens: $2.00/1M input, $0.500/1M cached, $6.00/1M output per 1M. At or above 200K, the whole request uses the long-context band.
What is the Grok 4.7 API model id?
Use grok-4.7 on the xAI API.
Does Grok 4.7 have a long-context surcharge?
Yes. At or above 200K prompt tokens, the entire request bills at higher input, cache, and output rates.
Is Grok 4.7 cheaper than GPT-6 Sol?
On output stickers under 200K, yes ($6 vs $10). Input stickers match at $2. Cache reads favor Sol ($0.20 vs $0.50). Measured task cost can still favor either side.
Is Grok 4.7 different from Grok 4.6 on price?
Under 200K, this catalog stores the same list rates. Prefer 4.7 for new agent and coding workloads per the model summary.
Are browser token counts Exact for Grok?
No. TokenCalculator labels Grok Approx. Confirm shipping budgets on xAI usage.
Why is Grok cheaper per token but sometimes more expensive per task?
If Grok emits more output tokens or takes more agent steps, the cheaper output sticker still loses. Price finished tasks, not stickers alone.
Are these rates live?
No. Curated list prices with lastVerified 23 Sep 2026. Confirm on xAI docs.
Next steps
Related Sep 2026 pricing guides
Frontier rate cards, cost-per-task math, and long-context cliffs. Cross-link these when you compare Sol, Luna, Opus 5.5, Grok 4.7, or Astra.