The rate card
Released September 21, 2026, Grok 4.7 is xAI's frontier model for coding, agentic tasks, and knowledge work, with a 500K-token context window and adjustable reasoning effort (low, medium, high, xHigh — high is the default). Pricing is unchanged from Grok 4.6, per xAI's release notes — verified October 28, 2026.
| Prompt size | Input / 1M | Cached input / 1M | Output / 1M |
|---|---|---|---|
| Under 200K prompt tokens | $2.00 | $0.50 | $6.00 |
| 200K+ prompt tokensWhole request billed at the higher tier | $4.00 | $1.00 | $12.00 |
The threshold is per-request, based on prompt tokens — not an account setting. A 199K-token prompt and a 200K-token prompt can have meaningfully different bills, because the higher rate applies to all tokens in that request, not just the tokens past the line. xAI's regional documentation prices regional endpoints at exactly +10% above global rates.
What the 200K cliff does to real jobs
250K input + 25K output (over the cliff): 0.25 × $4 + 0.025 × $12 = $1.30
For long agent sessions, the cliff — not the headline rate — is the number to watch. Agentic coding harnesses with growing context routinely cross it, which is one reason reported real-world harness costs vary so widely for the same model.
Where Grok 4.7 sits on price
At $2/$6, Grok 4.7 matches GPT-6.1 Sol and Claude Sonnet 5 on input while charging 40% less than Sol on output — so it gets cheaper the more output a workload produces. It is priced well under Claude Opus 5.5 ($4/$20) and GPT-6 Astra ($10/$50), and well above DeepSeek V4.1 Flash ($0.15/$0.60 off-peak). Detailed head-to-head math: Grok 4.7 vs GPT-6.1 Sol.
A note on performance claims
xAI's launch materials report Grok 4.7 at 37.6% on Terminal-Bench 4.0 (xAI's own harness, marked as a high-effort score) against 20.3% for Grok 4.6 — but independent measurement by Artificial Analysis put the same model at roughly 26% on the same benchmark. Vendor benchmarks and independent replication disagree here; treat xAI's figures as vendor-published, not independently verified. JetAI publishes no in-house benchmark scores, so we make no JetAI-measured quality claims for this model. Compare on price, then test on your own workload.
When Grok 4.7 makes sense
- Output-heavy workloads where the $6/M output rate (vs $10 on Sol) dominates the bill
- Prompts reliably under 200K tokens, so the cliff never triggers
- Agentic coding loops that benefit from explicit reasoning-effort controls
And when prompts are long or budgets are tight, check DeepSeek's flat long-context rates — and always model the cliff with our bill estimator.
Quick answers
How much does Grok 4.7 cost?
Verified October 28, 2026 per xAI's release notes: $2 per million input tokens, $0.50 cached input, and $6 per million output tokens for prompts under 200K tokens. Prompts of 200K tokens or more are billed at $4/$1/$12 — the higher rate applies to the entire request, not just the excess.
What is the Grok 4.7 context window?
500,000 tokens, per xAI's release notes and model documentation. Note the pricing cliff inside that window: at 200K prompt tokens the per-token rates double for the whole request, so a 250K-token prompt costs proportionally far more than a 150K-token one. Window size and billing band are separate numbers — budget against the band.
Is Grok 4.7 cheaper than GPT-6.1 Sol?
On input they match at $2/1M. On output, Grok 4.7 charges $6/1M against GPT-6.1 Sol's $10 — 40% less. But Sol offers a 50% Batch API discount and cheaper cached input ($0.10 vs $0.50), so the cheaper model depends on the workload shape. See our head-to-head comparison for the math.
Does Grok 4.7 have a batch API discount?
Not on the current public rate card — xAI lists batch discounts only for the older 4.3/4.20 family, not for Grok 4.7. The available cost levers are cached input at $0.50/1M, reasoning-effort settings that trim output tokens, and keeping prompts under the 200K-token pricing cliff where rates double.