xAI's Grok 4.3 now available on Amazon Bedrock

xAI and Amazon announced that Grok 4.3, a large language model built for enterprise reasoning, is now generally available on Amazon Bedrock. The launch is co-authored with Eric Jiang from xAI. Grok 4.3 is designed to give teams building agents and AI workflows a model that reasons reliably over long inputs.

The model includes several capabilities tailored to enterprise and agentic workloads. It accepts both text and image input and has a 1 million token context window to handle long documents and multi-turn conversations. One key feature is configurable reasoning effort: developers can set the reasoning level to none, low, medium, or high on a per-request basis, allowing a single model to serve both latency-sensitive calls (like classification at none effort) and complex reasoning tasks (like contract analysis at high effort). Grok 4.3 handles tool calling and instruction following well, making it practical for agents that depend on function calls, and it runs on Mantle, Amazon Bedrock's next-generation inference engine.

On xAI's benchmarks at the time of model launch, Grok 4.3 ranked #1 on the Artificial Analysis Omniscience benchmark with the lowest hallucination rate among frontier models compared. It also ranked #1 on the Artificial Analysis Tau2 Telecom benchmark for tool calling in customer support scenarios, and #1 on the Vals AI Case Law and Corporate Finance benchmarks for document understanding. xAI reports the model is positioned on the intelligence-versus-cost Pareto frontier, offering 2 to 10 times more intelligence per dollar than other frontier models.

Developers access Grok 4.3 through OpenAI-compatible APIs on Mantle. The endpoint URL is region-specific; for example, in us-west-2 the base URL is https://bedrock-mantle.us-west-2.api.aws/openai/v1. Developers can invoke the model using the OpenAI SDK or direct HTTPS requests to the Chat Completions API or Responses API. Authentication options include short-term bearer tokens generated from IAM credentials (recommended for production) or long-term Amazon Bedrock API keys (for exploration and getting started).

The model's default parameters differ slightly from OpenAI's standard: temperature defaults to 0.7 rather than 1, top_p defaults to 0.95 rather than 1, and max_completion_tokens defaults to 131072. When using the Responses API to access reasoning output, developers can set the effort level on each request and choose between stateful (where the service retains reasoning across turns) or stateless (where encrypted reasoning is passed back in the next request) patterns. Grok 4.3 also supports structured output via JSON Schema in strict mode and standard OpenAI-compatible tool calling, where developers describe available tools and the model decides whether and when to call them, returning a structured request that code can execute and feed back into the conversation. These features are designed for practical use cases such as contract review, credit agreement analysis, and financial document question answering, where the model reasons over long inputs and then calls out to systems of record.

Grok 4.3's launch on Amazon Bedrock marks xAI's entry as a model provider on the platform and reflects a broader trend of specialized reasoning models gaining access to enterprise cloud infrastructure. The model's design—with configurable reasoning effort, strong tool use, and document-handling capability—addresses a specific set of enterprise workloads: contract review, credit analysis, case law research, and financial document Q&A, where accuracy and the ability to reason over long inputs matter more than latency alone.

The positioning around cost efficiency (2 to 10 times more intelligence per dollar than frontier competitors) and benchmark leadership on hallucination rate and tool-calling accuracy suggests xAI is competing not on raw capability claims but on practical reliability and operational efficiency. This appeals to development teams building production agents that must make reliable function calls and reason correctly over legal and financial documents, where errors carry real cost.

The technical implementation via Mantle and OpenAI-compatible APIs lowers the friction for adoption; teams already using OpenAI SDKs can swap in Grok 4.3 with minimal code changes. The stateful multi-turn pattern with reasoning traces retained server-side, or the stateless encrypted reasoning option, offers flexibility for different deployment patterns and data residency requirements.

No comments yet. Be the first to share your thoughts!