NEWEST GROK500K CONTEXTTEXT + IMAGE INPUTREASONING TO XHIGH2x ABOVE 200K PROMPT20% OFF ON BATCH
Grok 4.7 API Pricing
Grok 4.7 is SpaceXSI's newest frontier model for coding, agentic tasks, and knowledge work, released September 21, 2026 on a new, larger base model than Grok 4.6 — on an identical rate card. The pricing page lists $2.00/M input, $0.50/M cached input, and $6.00/M output for prompts under 200K tokens. At or above 200K every rate doubles — $4.00 / $1.00 / $12.00 — and the higher rate applies to the whole request, not just the tokens past the line.
Input - per 1M tokens
$2.00/M
Under 200K prompt live
Output - per 1M tokens
$6.00/M
Reasoning billed as output live
Cached input
$0.50/M
Prompt cache SpaceXSI row -75%
Effective - agentic blend
$1.19/M
92/8 split - 82% cache
§ 01 / TERMINAL
Grok 4.7 API cost calculator.
Live calculator pre-loaded with Grok 4.7's sub-200K rates from SpaceXSI's pricing page. Reasoning effort is configurable across low, medium, high (default) and xhigh, and reasoning tokens bill at the output rate. Note the threshold: send a prompt of 200K tokens or more and every rate in this calculator doubles.
$ /mo
Workload split
Prompt cache hit rate
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
§ 02 / SCENARIOS
Grok 4.7 cost on real workloads.
AGENT
Tool-calling agent
$0.110/task
LONG CONTEXT
Research pack synthesis
$0.199/pack
CHAT
Power-user chat
$0.067/turn
RAG
Knowledge base RAG
$0.188/answer
Estimate · grok-tokenizer-estimate · ≈3.85 chars/token Auto-counts as you type
This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.
Characters 600
Words 101
Tokens (estimated) 156 tokens
Cost as input · uncached $0.00031 USD
Cost as output · uncached $0.00094 USD
Cost as cached input $0.00008 USD
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| Grok 4.7 Current | $2.00 cache $0.50 | $6.00 | $1.19 agentic 92/8 | 500K | Coding, agents, and knowledge work |
| Grok 4.6 | $2.00 cache $0.50 | $6.00 | $1.19 same | 500K | Same card on the previous base model |
| Grok 4.5 | $2.00 cache $0.30 | $6.00 | $1.04 −13% | 500K | Cache-heavy loops on an older Grok |
| Grok 4.3 | $1.25 cache $0.20 | $2.50 | $0.558 −2.1× | 1M | Lower-cost Grok with a 1M window |
| Grok Build 0.1 | $1.00 cache $0.20 | $2.00 | $0.476 −2.5× | 256K | Dedicated coding workflows |
| Claude Sonnet 5.5 | $2.00 cache $0.10 | $10.00 | $1.29 +9% | 1M | Claude-class coding agents |
| GPT-5.6 Sol | $4.00 cache $0.40 | $20.00 | $2.73 +2.3× · promo | 1.05M | OpenAI frontier workloads |
Grok 4.7 has published scores on 7 benchmarks we track; its best placing is #2 of 7 on Harvey Legal Agent Benchmark. 4 of these scores come from an independent leaderboard; the rest are the lab's own reports.
| Benchmark | Score | Rank | Leader | Source |
|---|---|---|---|---|
| Harvey Legal Agent Benchmark Professional work | 19.6% xhigh effort | #2 of 7 | Kimi K3.1 Pro · 23.7% | Vendor-reported SpaceXSI (xAI) |
| ARC-AGI-2 Reasoning | 61.4% ARC Prize verified, best result | #31 of 105 | GPT-6 Astra · 95% | Official leaderboard ARC Prize Foundation |
| ARC-AGI-1 Reasoning | 90.2% ARC Prize verified, best result | #33 of 99 | 6 tied · 98.5% | Official leaderboard ARC Prize Foundation |
| DeepSWE Coding | 71% high effort | #10 of 27 | Gemini 4 Argon · 77.9% | Vendor-reported SpaceXSI (xAI) |
| ARC-AGI-3 Reasoning · Standard harness | 1.8% ARC Prize verified, best result | #10 of 24 | GPT-6 Astra · 62.7% | Official leaderboard ARC Prize Foundation |
| Terminal-Bench 4.0 Agents and computer use | 37.6% xhigh effort, Grok Build harness, 330 trials, ±3.5 pts (95% CI) | #16 of 34 | Kimi K3.1 Pro · 65.7% | Official leaderboard Terminal-Bench (Stanford, Laude Institute) |
| CursorBench 4.0 Coding | 46.3% xhigh effort | #5 of 7 | Claude Opus 5.5 · 57.8% | Vendor-reported SpaceXSI (xAI) |
Grok 4.7 API pricing FAQ.
Grok 4.7 pricing questions, with SpaceXSI's token rates separated from the long-context threshold, the fast variant and the regional endpoint.
Q · 01 What is Grok 4.7's API price? +
SpaceXSI lists
grok-4.7 at $2.00/M input, $0.50/M cached input, and $6.00/M output for prompts below 200K tokens. These are USD prices per 1M tokens on the global endpoint, and reasoning tokens bill at the output rate.Q · 02 Is Grok 4.7 more expensive than Grok 4.6? +
No. SpaceXSI launched it on Grok 4.6's card — input, cached input, output and the doubled long-context tier all match — and its announcement says 4.7 is served at the same price and speed as 4.6, on a new, larger base model. Under our standard blend (92% input, 82% cache hits) Grok 4.7 works out to
$1.19/M and Grok 4.6 to $1.19/M. The cheaper Grok on cache-heavy loops is Grok 4.5, whose cache read is $0.30/M and whose blend is $1.04/M.Q · 03 Why does Grok 4.7 have two sets of prices? +
SpaceXSI charges long-context requests at double the rate:
$4.00/M input, $1.00/M cached input, and $12.00/M output. The trigger is the prompt reaching 200K tokens, and the higher rate then applies to every token in that request — not only the tokens beyond 200K. A 199K-token prompt and a 201K-token prompt therefore differ in cost by roughly 2x, not by 1%.Q · 04 What is Grok 4.7 Fast, and can I buy it on the API? +
It is the same Grok 4.7 served on faster infrastructure, at twice the standard token rates:
$4.00 input, $1.00 cached input and $12.00 output per 1M below a 200K prompt, and $6.00 / $1.50 / $18.00 above it. It is sold only inside Cursor and SpaceXSI's Grok Build coding agent — it is not on the public xAI API, and Grok Build's free tier does not include it. Cursor bills its own fast variant through the Cursor plan.Q · 05 Does Grok 4.7 get a batch discount? +
Yes, 20%, since late October 2026. Between October 19 and 21 SpaceXSI added
grok-4.7 to its batch list without a release note, and the model page now reads Batch API: Supported: batch requests bill at $1.60/M input and $4.80/M output, 20% off every token type, and most finish within 24 hours. Until then the model page said Batch API: Not supported. Priority processing runs the other way: it bills at 2x standard rates for higher scheduling priority.Q · 06 Does the US regional endpoint cost more? +
Yes, 10%. Requests sent to
us.api.x.ai keep inference in the United States and bill at 1.1x the global rates — the pricing page prints $2.20 input, $0.55 cached input and $6.60 output per 1M for Grok 4.7 below 200K, and $4.40 / $1.10 / $13.20 above. As of September 2026 SpaceXSI serves only Grok 4.7 and Grok 4.6 there.Q · 07 What changed from Grok 4.6? +
The price did not. SpaceXSI says 4.7 is a new, larger base model trained with a longer reinforcement-learning run on harder, multi-hour tasks, and that it is better at checking its own work and at long context. Its own launch table has Grok 4.7 at 46.3% on CursorBench 4.0 and 38.0% on Terminal-Bench 4.0, against 40.4% and 20.3% for Grok 4.6 — but 4.7 was scored at xhigh effort and 4.6 at high, so part of that gap is effort rather than model, and none of it is independently verified. On the Responses API,
grok-4.7 always returns encrypted reasoning content.Q · 08 What context window and limits does Grok 4.7 have? +
The model page lists
500,000 tokens with text and image input, text output and no text output limit, and a knowledge cutoff of May 2026. Rate limits are 150 requests per second and 50M tokens per minute, in regions us-east-1, us-west-2 and us-central-1. Grok 4.3 and the Grok 4.20 snapshots keep the longer 1M window.