Grok 4.7 is now on Amazon Bedrock
SpaceXAI's new Grok 4.7 model is now available through Amazon Bedrock. The model is trained to perform better on multi-step, long-running tasks rather than single-turn questions. According to SpaceXAI, the model has learned to verify its own work and manage longer context.
Why it matters
Agents carry a task through many steps and must verify what they find along the way. Grok 4.7 forms a candidate answer, checks it against what it has learned, and refines it. The model exposes a reasoning effort setting that controls how much deliberation happens before each response.
Measured results
The AWS team ran a tool-calling agent on three tasks (diagnosing a two-service outage, scheduling maintenance under constraints, and choosing migrations under a budget) for 72 total runs. The agent answered 67 of the 72 runs correctly. At the highest reasoning level it answered all 18 correctly, while at low it answered 15 of 18 using about an eighth of the output tokens.
Access and technical details
Grok 4.7 runs on the bedrock-runtime endpoint through a cross-Region inference profile. The us.xai.grok-4.7 profile is for US data residency, while global.xai.grok-4.7 offers the widest capacity. The model supports the OpenAI-compatible Chat Completions and Responses APIs as well as the Amazon Bedrock Converse API. SpaceXAI lists a 500K token context window with text and image input and text output.
