xAI's Grok 4.3 Launches on Amazon Bedrock with 1M Token Context

xAI has made Grok 4.3 generally available on Amazon Bedrock, marking the company's entry as a Bedrock model provider. The model offers a 1 million token context window and configurable reasoning effort for enterprise AI workflows.

Technical Specifications

Grok 4.3 accepts text and image input and returns text output. The model runs on Mantle, Amazon Bedrock's next-generation inference engine, and uses OpenAI-compatible APIs rather than the standard Bedrock Runtime API. Pricing has not been disclosed.

The model includes four reasoning effort levels—none, low, medium, and high—configurable per request. According to xAI, this allows teams to use a single model for tasks ranging from low-latency classification (none effort) to complex contract analysis (high effort).

Default parameters differ from OpenAI's specification: temperature defaults to 0.7 (not 1.0), top_p defaults to 0.95 (not 1.0), and max_completion_tokens defaults to 131,072.

Benchmark Performance

According to xAI, Grok 4.3 ranked #1 on three benchmarks at launch:

• Artificial Analysis Omniscience benchmark with the lowest hallucination rate among compared frontier models
• Artificial Analysis Tau2 Telecom benchmark for tool calling in customer support
• Vals AI Case Law and Corporate Finance benchmarks for document understanding

xAI claims the model delivers "2 to 10 times more intelligence per dollar" than other frontier models on the intelligence-versus-cost Pareto frontier, though specific benchmark scores were not provided.

Access and Authentication

Developers access Grok 4.3 through region-specific Mantle endpoints (example: https://bedrock-mantle.us-west-2.api.aws/openai/v1) using either the OpenAI SDK or direct HTTPS requests. The model ID is xai.grok-4.3.

Two authentication methods are available: long-term Bedrock API keys for exploration, and AWS IAM-generated short-term bearer tokens for production use. Amazon recommends the token-based approach to avoid embedding long-lived credentials in applications.

Reasoning Configuration

The Responses API allows developers to control reasoning output. Higher effort levels generate more internal reasoning tokens, visible in the usage metadata. The service can retain reasoning traces server-side (with store=True) or return encrypted reasoning content for stateless workflows.

For example, a request with reasoning={"effort": "high"} will allocate more reasoning tokens to multi-step problems, while effort="none" disables reasoning entirely for latency-sensitive operations.

Target Use Cases

xAI positions Grok 4.3 for enterprise workloads requiring accuracy over long contexts: contract review, credit agreement analysis, financial document Q&A, and agentic workflows with tool calling. The 1 million token context window accommodates extended documents and multi-turn sessions.

What This Means

This launch gives AWS Bedrock customers access to a frontier reasoning model with explicit effort controls—a feature competing providers like Anthropic and OpenAI embed in their reasoning models but don't expose as a per-request parameter. The 1M context window matches Claude 3.5 Sonnet's capacity but comes with the flexibility to trade reasoning depth for speed on individual calls. For enterprises already on AWS infrastructure, Grok 4.3 provides another option for agentic workflows without introducing new vendor relationships, though the lack of disclosed pricing makes cost comparison impossible at launch.