Grok 4.3 Lands on Amazon Bedrock
xAI's frontier model is now GA on Amazon Bedrock with the same $1.25/$2.50 pricing, 1M context, and configurable reasoning — the cheapest tracked frontier model just became enterprise-deployable.
On June 17, xAI announced that Grok 4.3 is generally available on Amazon Bedrock — the first time an xAI frontier model has shipped through AWS's managed inference stack. The move matters because Grok was already the cheapest model in helloai's tracked set at $1.25 per million input tokens and $2.50 output, but many enterprise teams could not route production traffic to a provider outside their existing cloud contracts. Bedrock closes that gap without changing the rate card: the same per-token pricing that applies on docs.x.ai applies on Bedrock, alongside a one-million-token context window and configurable reasoning effort at none, low, medium, or high.
The technical integration runs through Bedrock Mantle, AWS's OpenAI-compatible inference endpoint. Developers call grok-4.3 via a standard responses API, passing a reasoning.effort parameter to scale compute to task difficulty. That configurability is not cosmetic. Grok 4.3's default behavior includes reasoning on every query through xAI's direct API; on Bedrock, teams can set effort to none for latency-sensitive lookups and reserve high effort for agentic chains that need deeper deliberation. For workloads where output tokens dominate cost — tool loops, multi-step retrieval, customer-support agents — the ability to dial reasoning down on simple turns is a direct bill reduction, not a benchmark curiosity.
xAI's launch post leans on enterprise agent benchmarks rather than arena rank. It claims the lowest hallucination rate among frontier models on Artificial Analysis Omniscience, top placement on Tau2 Telecom for real-world tool-calling in support scenarios, and leading scores on Vals AI legal and finance document tasks. Those claims are vendor-reported and should be treated as directional, but they align with how Grok is positioned on helloai: Honest Daily Use leader, optimized for high-stakes research where fabricated citations are worse than a slower answer. The Bedrock listing does not change Grok's 1490 Elo or its fourth-place rank in the tracked set — it changes where that capability can legally and operationally run.
The competitive picture shifts on procurement, not capability. Claude Opus 4.8 still leads coding at 1503 Elo and $5/$25. Gemini 3.1 Pro holds the reasoning crown at 1493. GPT-5.5 remains the enterprise agentic default for teams already on Azure or OpenAI contracts. Grok's Bedrock availability means AWS-native shops can now slot the Pareto-cheapest frontier option into the same IAM, VPC, and compliance envelope they use for Anthropic and Meta models — without accepting a second vendor relationship or routing sensitive prompts through xAI's consumer stack. For agent pipelines that burn millions of output tokens monthly, the twelve-to-one output price gap versus GPT-5.5 compounds faster than a thirteen-point Elo deficit.
What comes next is distribution, not a new model generation. Grok 4.4 remains unreleased; the Bedrock milestone is about enterprise reach for the current flagship. Teams evaluating Grok should test reasoning effort settings against their actual token profiles before assuming the headline $1.25/$2.50 rate matches production spend — high-effort agent runs still generate reasoning tokens. But for AWS shops that have been paying frontier prices because Bedrock was the only approved path, the tracked set's cost floor just became deployable without a procurement exception.