Grok 4.6 is available through Amazon Bedrock, and its model card lists a 500K-token context window, four reasoning-effort settings, and implicit prompt caching. For a Node.js 22 app, you can call it with the AWS SDK for JavaScript v3. Caching may reduce the cost and latency of repeated input, but it is best effort: arrange stable content at the beginning of the prompt, then check response usage to see whether cache reads occurred.
What Grok 4.6 on Bedrock offers
Amazon Web Services lists Grok 4.6 as an xAI model for coding, agentic tasks, and knowledge work. Its model card reports a 500K-token context window and configurable reasoning effort of low, medium, high, or xhigh. AWS records its launch date as August 18, 2026, lifecycle as active, and end of life as no sooner than August 18, 2027. These are model-card details, not independent performance test results. AWS Bedrock’s Grok 4.6 model card
As an Amazon Associate I earn from qualifying purchases.
Which Bedrock endpoint and model ID should you use?
The model ID depends on the Bedrock access path. On bedrock-mantle, the model ID is xai.grok-4.6. With bedrock-runtime, use a cross-Region inference profile: us.xai.grok-4.6 or global.xai.grok-4.6. AWS says Grok 4.6 is not available for in-Region inference on the runtime endpoint. Choose the profile appropriate to your geography and current account availability; the model card’s Converse example uses us.xai.grok-4.6. AWS Bedrock’s Grok 4.6 model card
This article’s code uses bedrock-runtime and the Converse API. If your application uses the OpenAI-compatible bedrock-mantle route, its request interface differs; do not substitute one endpoint’s model ID or request format for the other.
#1 Best Overall
How implicit prompt caching works
AWS lists Implicit Prompt Caching for Grok 4.6. With implicit caching, Bedrock and the model automatically attempt to reuse eligible prompt prefixes; the request does not specify explicit cache checkpoints. The model card does not list explicit prompt caching for Grok 4.6, so do not assume checkpoint syntax documented for other models applies here. Support and configuration can differ by model and API. AWS’s prompt caching guide
A cache hit is not guaranteed, even when requests repeat the same prompt. To make reuse more likely, place content that stays the same across requests first, followed by content that changes:
Rank #2
- Stable prefix: shared instructions, policies, and reference material that recur across calls.
- Variable tail: the user’s current question, data, or other request-specific content.
AWS describes caching as best effort and says an exact prefix match can improve the likelihood of reuse. Treat prompt layout as an opportunity for a cache hit, not a guarantee. AWS’s prompt caching guide
What caching changes about token efficiency and cost
Prompt caching can reduce latency and input-token costs when eligible repeated tokens are actually served from cache. It does not automatically lower every request’s cost, and it does not reduce output tokens. Bedrock charges cache-read tokens at a model-specific cache-read rate; uncached input remains subject to the standard input rate. AWS’s prompt caching guide
Rank #3
The following standard-tier per-million-token rates are those displayed on the Grok 4.6 model card as checked October 7, 2026. They are dated published rates, not a promise of permanent pricing.
| Routing category | Input per 1 million tokens | Output per 1 million tokens | Cache read per 1 million tokens |
|---|---|---|---|
| In-Region and Geo CRIS | $2.20 | $6.60 | $0.55 |
| Global CRIS | $2.00 | $6.00 | $0.50 |
The model card also lists Priority at 1.75× standard per-token pricing and Flex at 0.5×. The applicable route, service tier, and rates should be checked against the current model card before estimating spend. AWS Bedrock’s Grok 4.6 model card
Rank #4
For a real workload estimate, compare cache-read volume with total input tokens across representative repeated requests. Record the response’s cache-usage fields when exposed; prompt structure alone cannot tell you whether a hit occurred. AWS notes that cache usage is represented in response usage fields for supported implementations. AWS’s prompt caching guide
Call Grok 4.6 with Node.js 22 and AWS SDK v3
Node.js 22 is the runtime used in this example, not a Grok 4.6-specific requirement established by AWS. Use the current SDK and runtime compatibility guidance when configuring your project. The JavaScript SDK v3 exposes BedrockRuntimeClient and ConverseCommand; the Converse operation requires bedrock:InvokeModel permission. AWS SDK for JavaScript v3 ConverseCommand reference
Best Value
import { BedrockRuntimeClient, ConverseCommand } from "@aws-sdk/client-bedrock-runtime";
const client = new BedrockRuntimeClient({
region: process.env.AWS_REGION,
});
const command = new ConverseCommand({
modelId: "us.xai.grok-4.6",
messages: [
{
role: "user",
content: [{ text: "Explain prompt caching simply." }],
},
],
});
const response = await client.send(command);
console.log(response.output?.message?.content);This is an illustrative integration shape based on the AWS SDK reference and the model card’s Converse sample, not a tested code claim. Configure AWS credentials through the normal SDK credential chain, set AWS_REGION, and make sure the selected profile is available to your account. Use global.xai.grok-4.6 instead if that is the suitable documented profile for your route. The client call follows the SDK reference’s request pattern. AWS Bedrock’s Grok 4.6 model card AWS SDK for JavaScript v3 ConverseCommand reference
How to tell whether caching is helping
- Send representative requests with the stable material in the same leading order and the changing request content afterward.
- Inspect the response usage fields for cache-related usage where exposed by the supported implementation.
- Compare cache-read volume, total input, latency, and resulting charges across a realistic set of repeated requests.
A cache-read rate is not the same as a guaranteed discount on all input: only tokens actually served from cache use that rate. Measure your workload before projecting savings, and use the current model card for the route and tier you deploy. AWS’s prompt caching guide AWS Bedrock’s Grok 4.6 model card
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
