xAI Reaches Amazon's Enterprise Cloud
On 15 June 2026, xAI's Grok 4.3 became the first xAI model available on Amazon Bedrock, making xAI the third independent AI laboratory with a model on the platform — joining Anthropic and OpenAI. The launch was featured in the AWS Weekly Roundup on 22 June 2026, drawing attention from enterprise developers who have so far deployed Bedrock models primarily through Claude, GPT-4o, or the Bedrock-native Titan family. For engineering teams that run AI workloads on AWS, Grok 4.3 on Bedrock is now the most price-competitive frontier reasoning model on the platform, and it arrives with infrastructure that differs meaningfully from existing Bedrock model deployments.
What Grok 4.3 Brings to the Bedrock Model Catalogue
The pricing proposition is immediately striking. Grok 4.3 is priced at $1.25 per million input tokens and $2.50 per million output tokens at on-demand rates, with cached input available at $0.20 per million tokens. By comparison, the nearest competitive reasoning models on Bedrock carry substantially higher input costs. xAI and AWS describe Grok 4.3 as the cheapest US-lab frontier reasoning model on Bedrock by a wide margin — a claim with immediate relevance for high-volume agent workloads where inference costs accumulate quickly.
The context window is 1 million tokens, matching what Google Gemini models offer and substantially exceeding the context available on most Bedrock model cards. Maximum output length is 30,000 tokens per response. For development teams building agents over large documents, codebases, or extended conversation histories, this combination of a 1-million-token window and competitive pricing changes the economics of long-context tasks meaningfully.
The Mantle Inference Engine
Grok 4.3 on Bedrock runs on Mantle, a new inference engine that AWS developed specifically for price-performance on reasoning workloads. Mantle operates at the bedrock-mantle endpoint with an OpenAI-compatible API path, which means teams using OpenAI-compatible SDKs can route requests to Grok 4.3 with minimal code changes. Mantle supports tool calling, structured output, and response streaming — the three capabilities that enterprise agent frameworks rely on most heavily.
The use of Mantle rather than the standard bedrock-runtime endpoint is architecturally notable: AWS is treating Grok 4.3 as the first deployment of a new inference infrastructure tier, not simply adding another model to an existing catalogue. Its OpenAI-compatible path positions Mantle as a general-purpose high-efficiency inference layer rather than a one-off integration.
Configurable Reasoning and Multimodal Capabilities
Grok 4.3 offers four levels of reasoning effort: none, low, medium, and high. Developers can set reasoning effort at the request level, allowing cost and latency to be tuned per task — lightweight reasoning for simple classifications and full reasoning depth for complex multi-step agent tasks. This in-request configurability is particularly useful for multi-agent pipelines where different steps have different reasoning requirements.
Beyond text and code, Grok 4.3 accepts native video input — a capability not broadly available across the Bedrock model catalogue at launch. For teams building agents that process product videos, customer support recordings, or instructional content, native video input at Grok 4.3's pricing tier opens workloads that were previously too costly to run at scale.
How This Compares to Other Bedrock Models
Before Grok 4.3, enterprise developers choosing a reasoning model on Bedrock weighed Claude Opus 4.8 (strong performance, higher cost) against GPT-4o and o-series models (broadly capable, mid-tier pricing). Grok 4.3 enters at the low end of the cost spectrum with competitive reasoning performance and the highest context window on the platform. For cost-sensitive workloads — batch document processing, long-form analysis, high-volume agentic pipelines — Grok 4.3 changes which model a rational cost-optimising team would select.
What This Means for Indian Engineering Teams
For Indian IT services and product teams running AI inference on AWS, the Grok 4.3 pricing at $1.25 per million input tokens represents a concrete reduction in unit economics for high-volume inference workloads. Organisations building large-scale agent pipelines — customer service automation, document processing, or code review workflows — will find that Grok 4.3's combination of a 1-million-token context window and low on-demand pricing enables use cases that were previously cost-prohibitive.
For fintech and enterprise product teams in India processing large volumes of financial documents, legal contracts, or regulatory filings, the 1-million-token context means entire documents can be passed to a single agent session without chunking — reducing pipeline complexity, improving accuracy, and lowering the engineering overhead of managing multi-step retrieval workflows. The OpenAI-compatible Mantle path also reduces migration effort for teams already working with OpenAI SDK tooling.
The Bottom Line
On 15 June 2026, xAI's Grok 4.3 became available on Amazon Bedrock via the new Mantle inference engine, priced at $1.25 per million input tokens and $2.50 per million output tokens — the lowest on-demand price for a US-lab frontier reasoning model on the platform. With a 1-million-token context window, four configurable reasoning levels, native video input, and an OpenAI-compatible API path, Grok 4.3 is a credible cost-optimisation option for Indian engineering teams running high-volume agent workloads on AWS infrastructure.
Frequently Asked Questions
What is Grok 4.3 and when did it arrive on Amazon Bedrock?+
Grok 4.3 is xAI's frontier reasoning model, which became available on Amazon Bedrock on 15 June 2026. It is the first xAI model offered on the platform, making xAI the third independent AI laboratory on Bedrock alongside Anthropic and OpenAI. It features a 1-million-token context window, native video input, configurable reasoning effort levels (none, low, medium, and high), and support for tool calling, structured output, and response streaming. It runs on Mantle, a new AWS inference engine designed for price-performance on reasoning workloads.
How is Grok 4.3 priced on Amazon Bedrock compared to other models?+
Grok 4.3 is priced at $1.25 per million input tokens and $2.50 per million output tokens at on-demand rates, with cached input available at $0.20 per million tokens. AWS and xAI describe it as the cheapest US-lab frontier reasoning model on Bedrock by a wide margin at launch, positioned significantly below Claude Opus 4.8 and comparable OpenAI models on the platform. This pricing makes Grok 4.3 particularly attractive for high-volume agent workloads where inference costs accumulate rapidly.
What is the Mantle inference engine and how does it differ from standard Bedrock?+
Mantle is a new inference engine AWS built specifically for price-performance on reasoning workloads. Grok 4.3 runs at the bedrock-mantle endpoint rather than the standard bedrock-runtime endpoint, with an OpenAI-compatible API path. This means teams using OpenAI-compatible SDKs can route requests to Grok 4.3 with minimal code changes. Mantle supports tool calling, structured output, and response streaming, and represents a new infrastructure tier within Bedrock rather than a standard model catalogue addition.
What is the context window for Grok 4.3 and why does it matter for enterprise workloads?+
Grok 4.3 accepts up to 1 million tokens of input and generates responses up to 30,000 tokens long. A 1-million-token context window means teams can pass entire large documents — lengthy contracts, full codebases, extended conversation histories — to a single agent session without chunking the input into smaller pieces. Combined with Grok 4.3's low on-demand pricing, this makes long-context tasks economically viable at production scale in a way that shorter-context, higher-priced alternatives are not.
Written by
TechPillow Team
Sharing insights on technology, product development, and the Indian tech ecosystem.