Grok 4.7 is now accessible on Amazon Bedrock

xAI’s Grok 4.7 is now accessible on Amazon Bedrock, including a frontier mannequin constructed for coding, long-running brokers, and data work to the Bedrock mannequin catalog. It gives a 500K token context window and helps configurable reasoning effort at 4 ranges: low, medium, excessive, and xhigh.

Grok 4.7 is served on the bedrock-runtime endpoint by way of cross-Area inference profiles, and it helps the Responses, Chat Completions, and Converse APIs. In accordance with xAI, it’s their most succesful mannequin for coding and data work: it really works longer on tough duties and verifies its personal output extra fastidiously earlier than shifting on.

This publish covers what xAI says Grok 4.7 is designed for, how it’s packaged on Amazon Bedrock, and how you can ship your first request.

What Grok 4.7 is constructed for

The potential and coaching particulars on this part come from xAI’s launch announcement, Introducing Grok 4.7, revealed September 21, 2026, and from the Grok 4.7 model documentation.

xAI positions Grok 4.7 as its most succesful mannequin for coding and data work. The theme is endurance relatively than uncooked pace: the mannequin works longer on tough duties and checks its personal work extra fastidiously earlier than shifting on.

On coaching, xAI stories that Grok 4.7 makes use of a brand new and bigger base mannequin. It was skilled with an extended reinforcement studying run over a tougher mixture of duties, intentionally weighted towards issues that take many hours to finish. Two capabilities got here out of that: the mannequin is best at verifying its personal work, and makes more practical use of its 500K token context window on lengthy duties. xAI additionally skilled it to natively perceive the Grok Bot harness, which it credit for enhancements in conversational duties and basic data work.

For anybody constructing brokers, the self-verification conduct is the element price noting. A mannequin that checks its personal output earlier than persevering with tends to fail much less catastrophically on lengthy trajectories, the place an early mistake in any other case compounds by way of each later step.

xAI additionally calls out stronger doc and presentation technology, and describes good points on skilled data work of the type completed by attorneys, nurses, and monetary analysts. xAI stories good points throughout its revealed evaluations. These span software program engineering with CursorBench and DeepSWE, multi-hour terminal and workplace work with Terminal-Bench and AA Briefcase, electrical engineering with EEBench, authorized work with the Harvey Authorized Agent Benchmark, and scientific reasoning with HealthBench Skilled. For the outcomes themselves, see xAI’s announcement.

Impartial analysis

Synthetic Evaluation runs its personal evaluations relatively than counting on developer-reported figures, which makes it a helpful second reference alongside xAI’s revealed outcomes.

In accordance with Artificial Analysis, Grok 4.7 improves throughout its analysis suite. The biggest good points are on long-horizon agentic data work and on coding brokers run in xAI’s personal harness. The Intelligence Index is a composite constructed from a number of unbiased evaluations protecting agentic device use, reasoning and data, data reliability, and long-context work.

Grok 4.7 is measured at xhigh reasoning effort, and Grok 4.6 on the effort degree Synthetic Evaluation reported for every measure. The ultimate row is the tradeoff to plan for: the good points include roughly double the output tokens per process. That’s why it pays to set the trouble degree intentionally relatively than inheriting the default.

Security and cyber safety

In accordance with xAI, Grok 4.7 was constructed with a completely new safeguard stack and is the strongest mannequin it has examined on refusals and jailbreak resistance. xAI frames the purpose in dual-use domains reminiscent of cyber safety and organic work as holding two issues without delay: remaining helpful for professional duties whereas refusing harmful ones.

On cyber safety particularly, xAI stories the mannequin permits solely a small fraction of dangerous dual-use prompts by way of whereas not often blocking professional safety work. xAI has additionally begun giving chosen cyber safety companions invite-only entry to Grok 4.7’s red-team capabilities for protection analysis.

How Grok 4.7 is packaged on Amazon Bedrock

Grok 4.7 accepts textual content and picture enter and returns textual content. The mannequin is served on the bedrock-runtime endpoint by way of cross-Area inference profiles, so requests title a profile relatively than a naked mannequin ID:

Grok 4.7 helps the Responses API, the Chat Completions API, InvokeModel and the Converse API.

As a result of the mannequin is OpenAI-compatible, you will have a alternative of consumer. The OpenAI SDK works towards the /openai/v1 path with a bearer token, which may be both an Amazon Bedrock API key or a short-term token minted out of your AWS Identification and Entry Administration (IAM) credentials. The AWS SDKs attain the identical mannequin by way of Converse, signing requests along with your unusual AWS credentials. Use the OpenAI SDK in case you’re porting an current integration. Use Converse in order for you one message form throughout the fashions in your account, together with invocation logging and response streaming by way of the usual Bedrock occasion varieties.

Utilizing Grok 4.7 with Bedrock options

Implicit immediate caching applies mechanically to repeated immediate prefixes, so brokers that resend a big system immediate or reference doc on each flip pay the cached fee for that prefix. Amazon Bedrock Guardrails connect by ID and model on the request, making use of content material filters, denied subjects, personally identifiable info (PII) redaction, and phrase insurance policies to each the immediate and the response. That is helpful for a mannequin which may run unattended throughout many steps. With structured outputs, you possibly can constrain a response to a JSON Schema so downstream code can parse it immediately. Invocation logging captures every name in Amazon CloudWatch with the request, the response, and token counts together with reasoning tokens. This provides you an audit path for lengthy agent runs.

Areas and inference choices

Grok 4.7 routes by way of one in all two cross-Area inference profiles relatively than pinning to a single Area.

The International profile, international.xai.grok-4.7, routes every request to any supported business AWS Area, spreading load throughout extra capability, and is priced under a geographic profile. The tradeoff is much less management over the place a given request is served, which might imply extra variable latency.

The US geographic profile, us.xai.grok-4.7, retains processing inside the US geography, which addresses US knowledge residency necessities. Select it when you will have residency constraints or latency-sensitive visitors, and International when value and throughput matter extra.

Service tier and pricing

Normal is pay-per-token with no dedication, chosen by setting "service_tier": "default" or omitting the sector. Precedence delivers quicker, prioritized processing for a premium ("service_tier": "precedence"). Flex gives lower-cost entry for work that isn’t time-sensitive ("service_tier": "flex").

Service tier is a major value lever that you simply management. For per-token pricing throughout the tiers, see the Amazon Bedrock pricing web page.

Ship your first request

Earlier than your first name, verify that the mannequin is accessible to you within the Bedrock console for the AWS Area that you simply plan to make use of.

Set up the OpenAI SDK, and boto3 in case you plan to make use of the Converse API:

pip set up openai
pip set up boto3

Generate a long-term Amazon Bedrock API key from the Amazon Bedrock console for exploration, then set your atmosphere:

export OPENAI_API_KEY="<present your Bedrock API key>"
export OPENAI_BASE_URL="https://bedrock-runtime.us-east-1.amazonaws.com/openai/v1"

A primary request with the Chat Completions API:

from openai import OpenAI

consumer = OpenAI()

response = consumer.chat.completions.create(
    mannequin="us.xai.grok-4.7",
    messages=[
        {"role": "user", "content": "Can you explain the features of Amazon Bedrock?"}
    ],
)
print(response.selections[0].message.content material)

The identical name by way of the Responses API:

response = consumer.responses.create(
    mannequin="us.xai.grok-4.7",
    enter="Are you able to clarify the options of Amazon Bedrock?",
)
print(response.output_text)

And thru the Converse API with boto3. As a result of reasoning is at all times energetic, the primary content material block carries the reasoning and the reply sits in a later block, so search the blocks for the textual content relatively than indexing content material[0]:

import boto3

consumer = boto3.consumer("bedrock-runtime", region_name="us-east-1")

response = consumer.converse(
    modelId="us.xai.grok-4.7",
    messages=[
        {"role": "user", "content": [{"text": "Can you explain the features of Amazon Bedrock?"}]}
    ],
    inferenceConfig={"maxTokens": 2048},
)

blocks = response["output"]["message"]["content"]
textual content = subsequent(b["text"] for b in blocks if "textual content" in b)
print(textual content)

On Converse, you set the trouble degree by way of additionalModelRequestFields relatively than a reasoning parameter:

response = consumer.converse(
    modelId="us.xai.grok-4.7",
    messages=[{"role": "user", "content": [{"text": "What is 17*23? Number only."}]}],
    inferenceConfig={"maxTokens": 3000},
    additionalModelRequestFields={"reasoning_effort": "xhigh"},
)

Three operational notes. First, requests should title us.xai.grok-4.7 or international.xai.grok-4.7.

Second, bedrock:InvokeModel is evaluated towards three assets: your account’s default challenge, the inference profile you title, and the underlying basis mannequin (FM). The inspiration mannequin Amazon Useful resource Identify (ARN) is wildcarded throughout Areas as a result of cross-Area profiles route exterior the calling Area. Bearer-token authentication moreover requires bedrock:CallWithBearerToken, which boto3 and Converse don’t want:

{
    "Model": "2012-10-17",
    "Assertion": [
        {
            "Effect": "Allow",
            "Action": "bedrock:InvokeModel",
            "Resource": [
                "arn:aws:bedrock:{region}:{account-id}:project/default",
                "arn:aws:bedrock:{region}:{account-id}:inference-profile/us.xai.grok-4.7",
                "arn:aws:bedrock:*::foundation-model/xai.grok-4.7"
            ]
        },
        {
            "Impact": "Enable",
            "Motion": "bedrock:CallWithBearerToken",
            "Useful resource": "*"
        }
    ]
}

Record each inference profile you intend to name. Profiles are scoped individually, so a coverage naming us.xai.grok-4.7 doesn’t cowl international.xai.grok-4.7.

Third, the 2 authentication mechanisms cowl totally different code paths. An Amazon Bedrock API key in OPENAI_API_KEY travels as a bearer token and authenticates the OpenAI-compatible calls. The boto3 Converse examples signal with SigV4 as an alternative, drawing in your unusual AWS credentials from the atmosphere, a profile, or a job. Configure each in case you intend to make use of Converse alongside the OpenAI-compatible APIs.

Deal with a long-term API key as an exploration-only credential. For manufacturing, use short-term bearer tokens generated out of your IAM credentials with the aws-bedrock-token-generator package deal, since they expire mechanically and maintain entry tied to your IAM identification.

Working with reasoning effort

Reasoning is energetic on Grok 4.7, and energy degree is a major management you will have over its value and latency. You configure it by way of the reasoning parameter on the Responses API with low, medium, excessive, or xhigh, and thru additionalModelRequestFields on Converse.

The default is excessive. That’s price setting explicitly relatively than inheriting, as a result of leaving it unset on latency-sensitive or high-volume calls will spend extra reasoning tokens than these calls want.

As a result of the mannequin is skilled to work longer and confirm its personal output, increased effort buys greater than additional deliberation on a single reply: it buys extra self-checking throughout a protracted process. Brief extraction and classification calls belong at low. Multi-step planning, lengthy agent trajectories, and work the place an early error propagates are the place excessive and xhigh earn their tokens.

Reasoning content material is encrypted. You may have it returned by passing the next on a Responses API request.embody: ["reasoning.encrypted_content"]
Then ship that content material again on subsequent turns to present the mannequin its personal prior reasoning as context in a multi-turn dialog. The Chat Completions API doesn’t return reasoning tokens.

from openai import OpenAI

consumer = OpenAI()  # OPENAI_BASE_URL factors on the bedrock-runtime endpoint

response = consumer.responses.create(
    mannequin="us.xai.grok-4.7",
    reasoning={"effort": "excessive"},
    embody=["reasoning.encrypted_content"],
    enter="Clarify quantum entanglement merely.",
)
print(response.output_text)

To search out the place extra reasoning stops paying for itself, benchmark the degrees towards your individual workload.

Get began

With Grok 4.7 in Amazon Bedrock, you get a 500K token context window, 4 reasoning effort ranges, picture enter, device calling, and Geo and International cross-Area routing. All of it’s reachable by way of the Responses, Chat Completions, and Converse APIs.

To begin constructing, assessment the Grok 4.7 mannequin card for the present Area record, characteristic matrix, and parameter particulars, and test the Amazon Bedrock pricing web page for token charges. Should you generated a long-term Amazon Bedrock API key for exploration, delete it from the Amazon Bedrock console if you end up completed. A standing credential you not want solely widens your account’s assault floor.

Sources

Concerning the authors