How much does Grok 4.6 cost?

Below 200K prompt tokens: $2 per million input, $0.50 cached input, $6 output. At 200K and above: $4, $1 and $12. A fast variant costs twice the listed price.

Does the higher Grok 4.6 price apply only to tokens above 200K?

No, and this is the trap. xAI's documentation states that requests whose prompt reaches the threshold are billed at the higher rate for all tokens in the request. Crossing 200K roughly doubles the cost of the entire call.

What is Grok 4.6's context window?

500K tokens — but the pricing tier changes at 200K, so the usable window and the affordable window are different numbers.

Grok 4.6 Hides a Pricing Cliff

A level coastal path ending abruptly at a sheer cliff edge above the sea — ModelsAugust 12, 2026·2 min read·By the Automation Squad Research
Grok 4.6 Hides a Pricing Cliff
It isn't the tokens over the line that cost more. It's all of them.

By Robert MacKelfresh

Founder, Automation Squad · August 12, 2026

The short answer

Grok 4.6 has a 500K token context window. Below 200K prompt tokens it costs $2 per million input, $0.50 cached and $6 output. At 200K and above those rates double to $4, $1 and $12 — and xAI's documentation states that requests reaching the threshold are billed at the higher rate for all tokens in the request, not just the ones past it.

Pricing table

The two tiers, the cliff, and how to stay the right side of it

The tier jump is the part that catches people. Read the note under the table before you plan any long-context workload.

Understand what the threshold actually does
xAI's docs state that requests whose prompt reaches the listed token threshold are billed at the higher rate for all tokens in the request. A prompt at 199,000 tokens and one at 201,000 tokens do not differ by 2,000 tokens of cost — the second costs roughly double, entirely.
Find out whether anything you run gets near 200K
Long documents, whole-repository context, accumulated agent conversation history. Agent loops are the dangerous case because context grows across turns without anyone deciding it should.
Put a hard ceiling on accumulated context in agent loops
If a loop appends history each turn, it will eventually cross 200K on its own. Cap it, summarise older turns, or split the work — the cliff arrives silently otherwise.
Price your realistic worst case, not your average
Averages hide cliffs by definition. Take your largest plausible request, apply the 200K-plus rates to every token in it, and see whether that number is acceptable.
Remember the fast variant doubles again
The fast variant costs twice the listed price. Fast plus over-200K is four times the headline rate.

Per million tokens	Under 200K prompt	200K and above
Input	$2.00	$4.00
Cached input	$0.50	$1.00
Output	$6.00	$12.00
Applies to	—	ALL tokens in the request
Context window	500K	500K
Fast variant	2× the listed price	2× the listed price

Take it with you

GROK 4.6 PRICING — mind the cliff
Source: https://automationsquad.com/news/grok-4-6-pricing-cliff/

PER MILLION TOKENS      UNDER 200K      200K AND ABOVE
  Input ................. $2.00 ........... $4.00
  Cached input .......... $0.50 ........... $1.00
  Output ................ $6.00 ........... $12.00

  Context window ........ 500K
  Fast variant .......... 2x the listed price (so 4x if also over 200K)

THE CLIFF — this is the bit
xAI's docs: "requests whose prompt reaches the listed token threshold are
billed at the higher rate for ALL tokens in the request."

A 199K prompt and a 201K prompt do not differ by 2K of cost.
The 201K one costs roughly DOUBLE. Entirely.

CHECKLIST
[ ] Does anything I run get near 200K prompt tokens?
      [ ] long documents
      [ ] whole-repo context
      [ ] agent loops accumulating conversation history  <- the silent one
[ ] Have I capped accumulated context in agent loops?
[ ] Have I priced my WORST case, not my average?
      (averages hide cliffs by definition)
[ ] Am I using the fast variant on top?

BENCHMARKS (vendor-reported)
  AA Intelligence Index 61 · CursorBench v3.2 69.9%
  DeepSWE v1.1 65.9% · FrontierCode v1.1 61.3%

xAI released Grok 4.6 on August 12, 2026 with a 500K token context window, available on day one in Cursor and Grok Build, through the xAI API, and via partners including OpenRouter, Vercel and Cloudflare.

The facts, from xAI's own model documentation: below 200K prompt tokens the model costs $2 per million input tokens, $0.50 per million cached input, and $6 per million output. At 200K prompt tokens and above, those become $4, $1 and $12. The documentation is explicit about how the threshold works — "requests whose prompt reaches the listed token threshold are billed at the higher rate for all tokens in the request." A separate fast variant costs twice the listed price. On benchmarks xAI reports an Artificial Analysis Intelligence Index score of 61, CursorBench v3.2 at 69.9%, DeepSWE v1.1 at 65.9% and FrontierCode v1.1 at 61.3%.

Automation Squad's take: the pricing structure is more interesting than the benchmarks, and almost nobody flagged it. Tiered pricing usually works like tax brackets — the higher rate applies to the portion above the line. Here it applies to everything, which turns a gradual cost curve into a step. The workload most exposed is the one least likely to notice: an agent loop that appends conversation history each turn will drift toward 200K on its own, with no single decision to blame, and the bill roughly doubles the moment it crosses. A 500K context window advertised alongside a cliff at 200K means the usable window and the affordable window are different numbers, and only one of them is on the marketing page.

Run this now: check whether anything you run gets near 200K prompt tokens — long documents, whole-repository context, or accumulated agent history. If you have an agent loop that grows its own context, put a hard ceiling on it or summarise older turns, because otherwise the cliff arrives without anyone choosing it. Then price your largest plausible request at the 200K-plus rates across every token, not your average request, since an average is precisely the statistic that hides a step change.

Questions people are asking

How much does Grok 4.6 cost?: Below 200K prompt tokens: $2 per million input, $0.50 cached input, $6 output. At 200K and above: $4, $1 and $12. A fast variant costs twice the listed price.
Does the higher Grok 4.6 price apply only to tokens above 200K?: No, and this is the trap. xAI's documentation states that requests whose prompt reaches the threshold are billed at the higher rate for all tokens in the request. Crossing 200K roughly doubles the cost of the entire call.
What is Grok 4.6's context window?: 500K tokens — but the pricing tier changes at 200K, so the usable window and the affordable window are different numbers.
Where can I use Grok 4.6?: It was available on day one in Cursor and Grok Build, through the xAI API, and via partners including OpenRouter, Vercel and Cloudflare.