AI サービスが一時的に利用できないため、復旧後に翻訳を補完します。
xAI’s Grok 4.7 is now available on Amazon Bedrock, adding a frontier model built for coding, long-running agents, and knowledge work to the Bedrock model catalog. It offers a 500K token context window and supports configurable reasoning effort at four levels: low, medium, high, and xhigh. Grok 4.7 is served on the bedrock-runtime endpoint through cross-Region inference profiles, and it supports the Responses, Chat Completions, and Converse APIs. According to xAI, it’s their most capable model for coding and knowledge work: it works longer on difficult tasks and verifies its own output more carefully before moving on. This post covers what xAI says Grok 4.7 is designed for, how it is packaged on Amazon Bedrock, and how to send your first request. What Grok 4.7 is built for The capability and training details in this section come from xAI’s launch announcement, Introducing Grok 4.7, published September 21, 2026, and from the Grok 4.7 model documentation. xAI positions Grok 4.7 as its most capable model for coding and knowledge work. The theme is endurance rather than raw speed: the model works longer on difficult tasks and checks its own work more carefully before moving on. On training, xAI reports that Grok 4.7 uses a new and larger base model. It was trained with a longer reinforcement learning run over a harder mix of tasks, deliberately weighted toward problems that take many hours to complete. Two capabilities came out of that: the model is better at verifying its own work, and makes more effective use of its 500K token context window on long tasks. xAI also trained it to natively understand the Grok Bot harness, which it credits for improvements in conversational tasks and general knowledge work. For anyone building agents, the self-verification behavior is the detail worth noting. A model that checks its own output before continuing tends to fail less catastrophically on long trajectories, where an early mistake otherwise compounds through every later step. xAI also calls out stronger document and presentation generation, and describes gains on professional knowledge work of the kind done by lawyers, nurses, and financial analysts. xAI reports gains across its published evaluations. These span software engineering with CursorBench and DeepSWE, multi-hour terminal and office work with Terminal-Bench and AA Briefcase, electrical engineering with EEBench, legal work with the Harvey Legal Agent Benchmark, and clinical reasoning with HealthBench Professional. For the results themselves, see xAI’s announcement. Independent evaluation Artificial Analysis runs its own evaluations rather than relying on developer-reported figures, which makes it a useful second reference alongside xAI’s published results. According to Artificial Analysis, Grok 4.7 improves across its evaluation suite. The largest gains are on long-horizon agentic knowledge work and on coding agents run in xAI’s own harness. The Intelligence Index is a composite built from multiple independent evaluations covering agentic tool use, reasoning and knowledge, knowledge reliability, and long-context work. Measure (Artificial Analysis) Grok 4.7 Grok 4.6 Intelligence Index 46 44 Coding Agent Index 56 47 AA-Briefcase, long-horizon knowledge work (Elo) 1,657 1,546 GDPval-AA, professional work products (Elo) 1,695 1,605 AA-Omniscience Index 32 30 AA-Omniscience hallucination rate 29% 34% Output tokens per Intelligence Index task ~81k ~38k Grok 4.7 is measured at xhigh reasoning effort, and Grok 4.6 at the effort level Artificial Analysis reported for each measure. The final row is the tradeoff to plan for: the gains come with roughly double the output tokens per task. That’s why it pays to set the effort level deliberately rather than inheriting the default. Safety and cyber security According to xAI, Grok 4.7 was built with an entirely new safeguard stack and is the strongest model it has tested on refusals and jailbreak resistance. xAI frames the goal in dual-use domains such as cyber security and biological work as holding two things at once: remaining useful for legitimate tasks while refusing dangerous ones. On cyber security specifically, xAI reports the model allows only a small fraction of risky dual-use prompts through while rarely blocking legitimate security work. xAI has also begun giving selected cyber security partners invite-only access to Grok 4.7’s red-team capabilities for defense research. How Grok 4.7 is packaged on Amazon Bedrock Grok 4.7 accepts text and image input and returns text. The model is served on the bedrock-runtime endpoint through cross-Region inference profiles, so requests name a profile rather than a bare model ID: Inference option Model ID Base URL Geo cross-Region us.xai.grok-4.7 https://bedrock-runtime.{region}.amazonaws.com/openai/v1 Global cross-Region global.xai.grok-4.7 https://bedrock-runtime.{region}.amazonaws.com/openai/v1 Grok 4.7 supports the Responses API, the Chat Completions API, InvokeModel and the Converse API. Because the model is OpenAI-compatible, you have a choice of client. The OpenAI SDK works against the /openai/v1 path with a bearer token, which can be either an Amazon Bedrock API key or a short-term token minted from your AWS Identity and Access Management (IAM) credentials. The AWS SDKs reach the same model through Converse, signing requests with your ordinary AWS credentials. Use the OpenAI SDK if you’re porting an existing integration. Use Converse if you want one message shape across the models in your account, along with invocation logging and response streaming through the standard Bedrock event types. Using Grok 4.7 with Bedrock features Implicit prompt caching applies automatically to repeated prompt prefixes, so agents that resend a large system prompt or reference document on every turn pay the cached rate for that prefix. Amazon Bedrock Guardrails attach by ID and version on the request, applying content filters, denied topics, personally identifiable information (PII) redaction, and word policies to both the prompt and the response. This is useful for a model that might run unattended across many steps. With structured outputs, you can constrain a response to a JSON Schema so downstream code can parse it directly. Invocation logging captures each call in Amazon CloudWatch with the request, the response, and token counts including reasoning tokens. This gives you an audit trail for long agent runs. Regions and inference options Grok 4.7 routes through one of two cross-Region inference profiles rather than pinning to a single Region. The Global profile, global.xai.grok-4.7, routes each request to any supported commercial AWS Region, spreading load across more capacity, and is priced below a geographic profile. The tradeoff is less control over where a given request is served, which can mean more variable latency. The US geographic profile, us.xai.grok-4.7, keeps processing within the US geography, which addresses US data residency requirements. Choose it when you have residency constraints or latency-sensitive traffic, and Global when cost and throughput matter more. Service tier and pricing Standard is pay-per-token with no commitment, selected by setting "service_tier": "default" or omitting the field. Priority delivers faster, prioritized processing for a premium ("service_tier": "priority"). Flex offers lower-cost access for work that isn’t time-sensitive ("service_tier": "flex"). Service tier is a significant cost lever that you control. For per-token pricing across the tiers, see the Amazon Bedrock pricing page. Send your first request Before your first call, confirm that the model is available to you in the Bedrock console for the AWS Region that you plan to use. Install the OpenAI SDK, and boto3 if you plan to use the Converse API: pip install openai pip install boto3 Generate a long-term Amazon Bedrock API key from the Amazon Bedrock console for exploration, then set your environment: export OPENAI_API_KEY="" export OPENAI_BASE_URL="https://bedrock-runtime.us-east-1.amazonaws.com/openai/v1" A first request with the Chat Completions API: from openai import OpenAI client = OpenAI() response = client.chat.completions.create( model="us.xai.grok-4.7", messages=[ {"role": "user", "content": "Can you explain the features of Amazon Bedrock?"} ], ) print(response.choices[0].message.content) The same call through the Responses API: response = client.responses.create( model="us.xai.grok-4.7", input="Can you explain the features of Amazon Bedrock?", ) print(response.output_text) And through the Converse API with boto3. Because reasoning is always active, the first content block carries the reasoning and the answer sits in a later block, so search the blocks for the text rather than indexing content[0]: import boto3 client = boto3.client("bedrock-runtime", region_name="us-east-1") response = client.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"] text = next(b["text"] for b in blocks if "text" in b) print(text) On Converse, you set the effort level through additionalModelRequestFields rather than a reasoning parameter: response = client.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 must name us.xai.grok-4.7 or global.xai.grok-4.7. Second, bedrock:InvokeModel is evaluated against three resources: your account’s default project, the inference profile you name, and the underlying foundation model (FM). The foundation model Amazon Resource Name (ARN) is wildcarded across Regions because cross-Region profiles route outside the calling Region. Bearer-token authentication additionally requires bedrock:CallWithBearerToken, which boto3 and Converse don’t need: { "Version": "2012-10-17", "Statement": [ { "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" ] }, { "Effect": "Allow", "Action": "bedrock:CallWithBearerToken", "Resource": "*" } ] } List every inference profile you plan to call. Profiles are scoped individually, so a policy naming us.xai.grok-4.7 does not cover global.xai.grok-4.7. Third, the two authentication mechanisms cover 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 sign with SigV4 instead, drawing on your ordinary AWS credentials from the environment, a profile, or a role. Configure both if you intend to use Converse alongside the OpenAI-compatible APIs. Treat a long-term API key as an exploration-only credential. For production, use short-term bearer tokens generated from your IAM credentials with the aws-bedrock-token-generator package, since they expire automatically and keep access tied to your IAM identity. Working with reasoning effort Reasoning is active on Grok 4.7, and effort level is a primary control you have over its cost and latency. You configure it through the reasoning parameter on the Responses API with low, medium, high, or xhigh, and through additionalModelRequestFields on Converse. The default is high. That’s worth setting explicitly rather than inheriting, because leaving it unset on latency-sensitive or high-volume calls will spend more reasoning tokens than those calls need. Because the model is trained to work longer and verify its own output, higher effort buys more than extra deliberation on a single answer: it buys [truncated for AI cost control]