Amazon Bedrock Grok 4.6 API, Model IDs, Regions and Pricing

On this page

Quick answer

Amazon Bedrock exposes Grok 4.6 through two contracts that should not be mixed:

  • bedrock-mantle uses model ID xai.grok-4.6 and the OpenAI-compatible URL https://bedrock-mantle.{region}.api.aws/openai/v1. AWS documents an in-Region path in us-west-2.
  • bedrock-runtime uses the Bedrock runtime base URL and a cross-Region inference ID: us.xai.grok-4.6 for US Geo or global.xai.grok-4.6 for Global routing.

Choose the endpoint from data-location, feature, IAM, availability, latency, and price requirements—not from URL familiarity. Recheck the live model card because regions, quotas, features, and prices can change.

Endpoint decision table

Decisionbedrock-mantlebedrock-runtime
Model namexai.grok-4.6us.xai.grok-4.6 or global.xai.grok-4.6
Routing at source checkIn-Region in us-west-2Geo or Global cross-Region
API styleOpenAI-compatible endpointBedrock Runtime OpenAI-compatible base URL; AWS also documents Responses, Chat Completions, and Converse support
Feature contractClient-side tools, reasoning, projects, streaming, structured outputs, prompt cachingInvocation logs, streaming, prompt caching, reasoning; server-side tools, structured outputs, count tokens, and application inference profiles are not listed
Key evidenceExact Region availabilityInference-profile IAM plus route geography

The feature matrix is endpoint-specific. “Bedrock supports Grok 4.6” does not mean every Bedrock API, tool mode, structured-output path, or Region is available.

Price and caching boundary

AWS’s August 20 model card lists Standard-tier prices per million tokens:

Inference optionInputOutputCache read
In-Region$2.20$6.60$0.55
Geo cross-Region$2.20$6.60$0.55
Global cross-Region$2.00$6.00$0.50

Treat this as a dated record. Estimate uncached input, cached input, output, retries, tool calls, long trajectories, logging, network, and failed requests. Verify that the prompt actually hits the cache rather than assuming a repeated prefix will be billed at the cache-read rate.

IAM and data-location checklist

Before the first production request:

  1. Record the endpoint hostname, caller Region, exact model or inference-profile ID, project, and AWS account.
  2. Decide whether the workload requires single-Region, geography-bounded, or Global routing; do not silently broaden the route on retry.
  3. Scope bedrock:InvokeModel to the required model or inference profile and the default project resources documented for the chosen path.
  4. Confirm invocation logging, retention, encryption, and access controls with test evidence.
  5. Query current quotas in the target account and Region; a documentation listing is not account capacity.
  6. Test reasoning parameters, tool calls, streaming, caching, errors, throttling, and rollback on the exact endpoint.

Use the Grok 4.6 deployment readiness checker to turn these choices into a preflight result. Use the direct Grok 4.6 model guide when AWS hosting is not a requirement.

Frequently asked questions

What is the Amazon Bedrock Grok 4.6 model ID?

AWS documents xai.grok-4.6 for bedrock-mantle, us.xai.grok-4.6 for US Geo cross-Region inference, and global.xai.grok-4.6 for Global cross-Region inference on bedrock-runtime.

Can bedrock-runtime run Grok 4.6 in one Region?

Not according to the August 20 model card. AWS lists Geo and Global inference IDs for bedrock-runtime; the documented in-Region path is bedrock-mantle in us-west-2. Recheck the live regional table before deployment.

How much does Grok 4.6 cost on Bedrock?

At the source check, AWS listed Standard-tier prices per million tokens of $2.20 input and $6.60 output for In-Region or Geo, and $2 input and $6 output for Global. Cache reads were $0.55 and $0.50 respectively. Recheck current price and account terms.

Official sources

Source check: August 20, 2026. Verify the live endpoint, Region, inference ID, feature matrix, IAM requirements, quotas, logging, price, and data-routing contract before production use.