Grok 4.6 API: Pricing, the 200K Cliff and Setup

Grok 4.6 is xAI's new flagship, released on August 12, 2026. The API model ID is grok-4.6, the context window is 500,000 tokens, and the list price is $2 per 1M input tokens and $6 per 1M output tokens.

That is the same headline price as Grok 4.5. The number that actually changed is the one most pricing tables leave out.

Grok 4.6 API pricing #

All figures are USD per 1M tokens. The first row is the standard rate for prompts under 200K tokens; the second is long-context pricing at or above 200K.

The cached-input rate went up 67% #

Grok 4.6 and Grok 4.5 are both $2 input and $6 output. Compare the cache column, though:

Cached input on 4.6 costs 67% more than on 4.5. For a one-off chat completion that is noise. For the workloads people actually put on Grok — coding agents replaying a large system prompt and file context on every turn — it is the dominant term in the bill.

Take a 100K-token prompt where 90% of it is a stable cached prefix, which is roughly the cache-hit ratio these agent loops reach in practice:

grok-4.5   10,000 fresh × $2.00/1M  +  90,000 cached × $0.30/1M  = $0.0470
grok-4.6   10,000 fresh × $2.00/1M  +  90,000 cached × $0.50/1M  = $0.0650

Same sticker price, 38% more per prompt on the input side. If you are upgrading a cache-heavy agent from 4.5 to 4.6, that is the line to model before you switch, and it is the reason "Grok 4.6 costs the same as 4.5" is not a safe assumption.

The 200K cliff still applies #

xAI's long-context rule is unchanged and it is not a marginal rate: *requests whose prompt reaches the listed token threshold are billed at the higher rate for all tokens in the request.* There is no blended calculation.

199,000-token prompt on grok-4.6  →  199,000 × $2.00/1M  = $0.398
201,000-token prompt on grok-4.6  →  201,000 × $4.00/1M  = $0.804

Two thousand extra tokens double the cost of the request. If you are anywhere near 200K, trimming the prompt below the threshold is worth more than any other optimisation available to you.

How Grok 4.6 compares on price #

Flagship list prices, USD per 1M tokens:

On a 40K-in / 2K-out agent turn with no cache, Grok 4.6 costs $0.092 against $0.25 for Claude Opus 4.8 and $0.26 for GPT-5.6 Sol. It is priced as a value flagship, not a premium one.

What it actually does #

We called grok-4.6 through RouterPlex before publishing this. Verified behaviour:

One practical note: Grok 4.6 is a reasoning model, and reasoning tokens are billed as output at $6 per 1M. A short reply is not necessarily a cheap one — a two-token answer preceded by 200 reasoning tokens bills as 202 output tokens. Check usage.completion_tokens_details.reasoning_tokens if a simple prompt costs more than you expected.

Call Grok 4.6 with RouterPlex #

Create a RouterPlex key, give it a hard budget, and send a standard chat-completions request:

curl https://api.routerplex.com/v1/chat/completions \
  -H "Authorization: Bearer $ROUTERPLEX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "grok-4.6",
    "messages": [
      {"role": "user", "content": "Name the riskiest assumption in this migration plan."}
    ]
  }'

The Python version uses the regular OpenAI client — only the base URL, key and model ID change:

import os
from openai import OpenAI
 
client = OpenAI(
    api_key=os.environ["ROUTERPLEX_API_KEY"],
    base_url="https://api.routerplex.com/v1",
)
 
response = client.chat.completions.create(
    model="grok-4.6",
    messages=[{"role": "user", "content": "Design a rollback-safe deployment plan."}],
)
 
print(response.choices[0].message.content)

Claude Code and the Anthropic SDK reach the same model through the Anthropic-compatible format — set ANTHROPIC_BASE_URL=https://api.routerplex.com and use grok-4.6 as the model. See the Claude Code setup guide for the full configuration.

Should you switch from Grok 4.5? #

• Cache-light workloads — one-off completions, short prompts, batch classification. Same price, newer model. Switch.
• Cache-heavy agent loops — a large stable system prompt replayed every turn. Price the 38% input increase from the worked example above against whatever quality gain you measure. It may still be worth it; it is not free.
• Anything near 200K tokens — the cliff behaves identically on both. Trim below the threshold first, then pick a model.

Both are live on RouterPlex at xAI list prices with no per-token markup, so you can A/B them on the same key and compare real spend. See the live pricing for grok-4.6 or the full Grok API pricing breakdown for the rest of the xAI catalogue.