Amazon Bedrock Adds Grok 4.7 With 500K Context Window
Amazon Bedrock Grok 4.7 access launched on September 28, bringing SpaceXAI’s newest coding and knowledge-work model into Amazon Web Services’ managed generative AI platform. The release gives enterprise developers another route to the model without integrating directly with SpaceXAI’s API.
The Bedrock addition arrives one week after SpaceXAI introduced Grok 4.7. AWS is presenting the model as a fit for long-running agents, software engineering and knowledge work, while keeping deployment inside the cloud controls many companies already use.
The Amazon Bedrock release adds four practical capabilities:
- A 500,000-token context window
- Low, medium, high and xhigh reasoning levels
- Text and image input
- Responses, Chat Completions and Converse API support
How Amazon Bedrock Serves Grok 4.7
AWS says Grok 4.7 runs through the Bedrock Runtime endpoint using cross-Region inference profiles. Developers can identify the model with either a United States or global inference profile, allowing Bedrock to route requests across supported capacity rather than tying every call to one region.
The model is available through three interface families. The OpenAI-compatible Responses and Chat Completions APIs can reduce migration work for teams already using those request formats, while AWS’s Converse API offers a common interface across different Bedrock models.
Amazon’s official model card lists Grok 4.7 as active, with text and image input and a 500,000-token context window. AWS also says the model will not reach end of life before September 28, 2027, and will receive a legacy period of at least six months.
Grok 4.7 Targets Long-Running Coding Tasks
SpaceXAI describes Grok 4.7 as a larger base model than Grok 4.6, trained with a longer reinforcement-learning run on a harder task mix. The company says the training emphasized problems that take many hours, along with stronger self-checking and better management of lengthy context.
That positioning matters because coding agents increasingly need to explore repositories, modify several files, run tests and correct failures across extended sessions. A large context window can help preserve specifications, code and intermediate results, although usable agent performance still depends on tool design, permissions and evaluation.
SpaceXAI reports a 46.3% score for Grok 4.7 xHigh on CursorBench 4.0, compared with 40.4% for Grok 4.6 High. It also reports 37.6% on Terminal-Bench 4.0, up from 20.3% for Grok 4.6. Those are vendor-published results and should be treated as directional until broader independent testing accumulates.
Pricing and Reasoning Controls
SpaceXAI prices standard Grok 4.7 API access from $2 per million input tokens and $6 per million output tokens, matching the standard rate it lists for Grok 4.6. A faster variant doubles output speed and carries a higher price, while long-context requests follow separate pricing rules.
Bedrock customers pay through AWS rather than the direct SpaceXAI console, so their effective bill also depends on Amazon’s service pricing and the selected deployment path. Teams should compare complete task cost, latency and output quality instead of assuming the lowest token rate produces the cheapest completed workflow.
The four reasoning levels give developers a direct trade-off. Low effort can suit latency-sensitive or routine work, while high and xhigh allow more computation for complex coding, analysis and agent steps. That control can also help organizations reserve the most expensive mode for tasks that justify it.
Bedrock Expands the Enterprise Route to Grok
Grok 4.7 was already available through SpaceXAI’s API, Cursor, Grok Build, model routers and other cloud platforms. The Bedrock release is therefore an access expansion, not the model’s original debut. Its significance comes from placing Grok beside other foundation models in an environment built around enterprise identity, monitoring and deployment controls.
For AWS customers, a common platform can simplify model comparison and reduce the operational cost of maintaining separate provider integrations. It also makes it easier to test Grok against models already available in Bedrock using the same application architecture and governance process.
SpaceXAI says Grok 4.7 uses a new safeguard stack and reports improved refusal and jailbreak resistance. Its launch materials cite a 3.3% risky-prompt allowance rate on the company’s HackerBench v0.3 evaluation. Customers should still apply their own access limits, logging and human approval requirements before granting any model authority over production systems.
The near-term test will be whether Grok’s longer-context and self-verification claims translate into fewer failed agent runs at production scale. Bedrock availability broadens the number of enterprises able to run that comparison inside an existing cloud environment, while AWS’s lifecycle commitment gives teams a clearer planning horizon for evaluation and deployment.
