xAI's Grok 4.3 launches on Amazon Bedrock with 1M context and always-on reasoning, first xAI model available through AWS

xAI's Grok 4.3 became available on Amazon Bedrock on June 15, 2026, marking the first time an xAI model has been offered through Amazon's managed AI service. The launch adds xAI to Bedrock's roster of third-party model providers alongside Anthropic, Meta, and Mistral, and introduces Mantle—a new inference engine in Bedrock with an OpenAI-compatible endpoint designed for price performance.

What's new

Grok 4.3 is available through the bedrock-mantle endpoint using the model ID xai.grok-4.3. Access is via the OpenAI-compatible Chat Completions and Responses APIs—not the standard bedrock-runtime or Converse endpoints—allowing developers to route existing OpenAI-API-compatible code to Grok 4.3 by changing a base URL and model identifier.

Key specifications:

• Context window: 1 million tokens
• Reasoning: Always-on, with configurable effort levels: none, low, medium, high
• Input modalities: Text, images
• Model ID: xai.grok-4.3 on the bedrock-mantle endpoint
• Endpoint URL: https://bedrock-mantle.{region}.api.aws/openai/v1
• Regions at launch: us-west-2 (Oregon), us-east-1 (N. Virginia), us-east-2 (Ohio)
• Service tiers: Standard, Priority, and Flex throughput options

The reasoning implementation is distinctive. Rather than an optional extended thinking mode, Grok 4.3 reasons by default. Developers configure effort via the reasoning parameter, and encrypted reasoning content can be returned and passed back in subsequent turns for multi-turn reasoning continuity. The Chat Completions API does not expose reasoning tokens; the Responses API does.

AWS documentation lists contract review, case law research, credit agreement analysis, and financial document Q&A as primary enterprise use cases.

Context

Amazon Bedrock has become the dominant enterprise AI marketplace for managed third-party model access. Before this launch, Bedrock carried Anthropic Claude, Meta Llama, Mistral, Cohere, and others—but no xAI model. Adding Grok 4.3 makes it accessible to the large population of AWS enterprise customers who standardize on Bedrock for IAM authentication, native CloudTrail audit logging, and VPC-based data residency controls, without requiring a separate direct xAI API agreement.

The Mantle inference engine is new Bedrock infrastructure: it provides an OpenAI-compatible path (openai/v1 via bedrock-mantle.{region}.api.aws) rather than the older InvokeModel or Converse patterns used by most other Bedrock models. This OpenAI-API-compatible surface positions Mantle as a likely home for additional models requiring the same interface.

Grok 4.3 had been available previously through xAI's own API and via X. The Bedrock launch extends enterprise reach under AWS billing and compliance infrastructure.

Why it matters

For enterprises on AWS, Bedrock availability removes a key procurement friction: a separate xAI API contract, independent usage metering, and a separate cost center. Grok 4.3 joins the existing Bedrock catalog under standard AWS billing—one bill, one IAM policy, one audit log.

The always-on reasoning architecture is a meaningful design choice. Models that treat reasoning as opt-in require prompt engineering or explicit mode switches to activate consistent step-by-step thinking. Grok 4.3's default-on reasoning—configurable down to none when latency matters—means agents get consistent multi-step reliability without additional developer overhead.

The Mantle endpoint's OpenAI-API compatibility also lowers migration cost for teams evaluating Grok as an alternative to GPT-4-class models. No SDK change required, only configuration.