xAI's Grok 4.3 lands on Amazon Bedrock: 1 million token context window, configurable reasoning for the enterprise

On 15/06/2026, xAI officially brought Grok 4.3 to Amazon Bedrock — the first time xAI has become an official model provider on the AWS Bedrock platform. This release is aimed squarely at enterprise workloads, with a context window of up to 1 million tokens and a configurable reasoning mechanism that can be tuned per task.

The event: Grok 4.3 arrives on AWS Bedrock

According to announcements from AWS and xAI, on 15/06/2026 Grok 4.3 became an xAI model officially available on Amazon Bedrock. This is a notable milestone because xAI had never previously been a model provider on AWS's managed model platform — now Bedrock customers can call Grok alongside the other models already available within the same service.

The model runs on a new Bedrock inference engine called "Mantle", accessed through an OpenAI-compatible endpoint, with the model ID xai.grok-4.3. At launch, the model is available in the us-west-2, us-east-1 and us-east-2 regions.

Key technical specifications

According to AWS documentation, Grok 4.3 on Bedrock has a context window of 1,000,000 tokens — large enough to load very long documents or many files at once. The default maximum output is 131,072 tokens.

Notably, the model is positioned as "reasoning-first", with a reasoning level that can be configured across four steps: none, low, medium, high (default low) — per AWS. This lets businesses trade off between reasoning depth and cost/latency depending on the task: simple tasks use a low level for speed, while complex tasks raise the reasoning level for a more thorough answer.

Aimed at enterprise workloads

This release is clearly positioned for enterprise tasks that require processing large volumes of text and multi-step reasoning. AWS cites representative scenarios including:

• Contract review — reading and cross-checking clauses within long documents.
• Case-law research — synthesizing and extracting from large volumes of legal text.
• Credit analysis — processing financial data and records.
• Q&A over financial documents — answering questions based on reports and long documents.

The 1 million token window combined with adjustable reasoning levels makes these "read a lot — reason deeply" tasks feasible on a single endpoint.

xAI's performance claims

It is important to clearly distinguish between technical specifications (published by AWS) and performance claims coming from the manufacturer itself. According to xAI's announcement, Grok 4.3 achieves the lowest "hallucination" (fabricated information) rate among frontier models and ranks #1 on the Artificial Analysis Omniscience leaderboard.

This is an xAI claim, not a neutral conclusion from an independent third party. As with any vendor-published benchmark figure, businesses should verify it on their own data and tasks before deciding, rather than relying entirely on leaderboards.

The Namtech view for Vietnamese businesses

Bringing a powerful model like Grok 4.3 to Bedrock opens up more options for businesses — particularly the ability to use multiple providers within the same AWS platform. However, the tasks AWS cites (contracts, case law, credit files, financial documents) all involve highly sensitive data.

For that kind of core data, a business that wants to use a powerful model while still keeping its data under its own control should consider an internal AI running on its own infrastructure (on-premise): the model runs locally, the data never leaves the organization, and there is no dependence on a foreign cloud service that could change its terms or access at any time.

FAQ

Note: This article was compiled from public sources as of 22/06/2026; specifications may change with AWS/xAI updates. The performance claims (hallucination, benchmarks) are statements from xAI, not independent evaluations.