Grok 4.7 Now Available on Amazon Bedrock for Long-Horizon Agents

TL;DR: SpaceXAI's Grok 4.7 is now available on Amazon Bedrock, featuring four reasoning effort levels (low, medium, high, xhigh) that control how much deliberation occurs before each response. AWS tested the model on three reasoning tasks across all effort levels, with the agent answering 67 of 72 runs correctly, including all 18 at xhigh, while the low setting used about one-eighth the output tokens.

What's New

Amazon Web Services has announced the availability of SpaceXAI's Grok 4.7 on Amazon Bedrock, its foundation model service. According to SpaceXAI, Grok 4.7 is trained to work longer on more difficult tasks and can verify its own work—capabilities particularly important for building agents that carry tasks through many steps.

Grok 4.7 introduces a configurable reasoning effort setting that controls how much deliberation happens before each response. The model supports four effort levels: low, medium, high, and xhigh. SpaceXAI reports that Grok 4.7 uses a larger base model than Grok 4.6 and was trained with a longer reinforcement learning run on a harder mix of tasks, weighted toward problems that take many hours to complete.

According to SpaceXAI's published benchmarks, the largest gains over Grok 4.6 appear in categories that measure sustained work rather than single-turn answers. For example, Terminal-Bench 4.0 (multi-hour terminal work) improved from 20.3% to 37.6%, and EEBench (electrical engineering) increased from 53.0% to 64.0%.

AWS tested Grok 4.7 by building a tool-calling agent and evaluating it on three reasoning tasks: diagnosing a two-service outage from alerts, scheduling maintenance under interacting constraints, and choosing an optimal set of migrations under a budget. Each task had exactly one correct answer verified by brute force. The agent ran all three tasks at every reasoning effort level six times each, for 72 total agent runs.

The results demonstrated a clear trade-off between accuracy and token usage. At xhigh effort, the agent answered all 18 runs correctly. At low effort, it answered 15 of 18 runs correctly while using approximately one-eighth of the output tokens. Overall, the agent answered 67 of the 72 runs correctly across all effort levels.

The model features a 500K token context window with text and image input and text output. On Amazon Bedrock, Grok 4.7 supports configurable reasoning effort, tool calling, and structured output. It's compatible with both the OpenAI-compatible Chat Completions and Responses APIs and the Amazon Bedrock Converse API.

Pricing and Availability

Grok 4.7 is accessible through cross-Region inference profiles on Amazon Bedrock. Two profiles are available: us.xai.grok-4.7 for US geography with data residency requirements, and global.xai.grok-4.7 for commercial AWS Regions without residency constraints. Cross-Region inference routes requests across multiple AWS Regions for higher throughput without managing capacity region by region.

AWS reports token counts rather than specific currency prices, as token prices change over time. For current token pricing, supported Regions, and service tiers, users should consult the Grok 4.7 model card in the Amazon Bedrock User Guide and the Amazon Bedrock pricing page.

How to Try It

To access Grok 4.7, users need an AWS account with access to Amazon Bedrock and model access enabled for Grok 4.7. Technical requirements include Python 3.9 or later with openai, boto3, and aws-bedrock-token-generator installed, plus AWS credentials configured in the environment.

The AWS Identity and Access Management (IAM) identity needs bedrock:InvokeModel permission on both the inference profile and the account's default project. AWS recommends generating short-term bearer tokens from existing AWS credentials for production use, so authentication stays tied to the IAM identity without storing long-lived secrets.

The model runs on the bedrock-runtime endpoint. When configuring reasoning tokens, users should note that they count as output tokens and draw down max_completion_tokens before answer text is produced. AWS used a limit of 32,768 tokens in testing, which provided adequate headroom at every effort level.

Who It's For

Grok 4.7 targets developers building long-horizon agents that require sustained reasoning over multiple steps. The configurable reasoning effort levels make it suitable for organizations that need to balance accuracy against token costs on a per-request basis.

The model is particularly relevant for applications involving multi-hour tasks, complex technical work, and scenarios where agents must verify their own work along the way. SpaceXAI's benchmark results suggest strong performance in electrical engineering, legal work, coding, clinical reasoning, and office automation tasks.

Alternatives

Amazon Bedrock offers foundation models from multiple providers, giving users alternatives to Grok 4.7. AWS blog post authors include Suheel Farooq, Anirban Gupta, Ikenna Izugbokwe, and James Wang. The announcement was published on October 13, 2026.