Grok 4.3 Lands on Amazon Bedrock With 1M Context and Lower Prices
xAI just did something it has never done before: it shipped a model on someone else's cloud. Grok 4.3 went generally available on Amazon Bedrock on June 15, 2026 — the first time any xAI model has lived outside xAI's own stack. For enterprises that were curious about Grok but allergic to wiring up a new vendor, the calculus just changed.
What actually shipped
Grok 4.3 arrives with a 1-million-token context window and a 30,000-token maximum output — enough to drop an entire mid-sized codebase or a stack of contracts into a single prompt without chunking gymnastics.
The pricing is the headline most teams will care about. On Bedrock, Grok 4.3 runs $1.25 per million input tokens and $2.50 per million output tokens, with cached input at $0.20 per million. That undercuts most frontier-tier models on the same platform.
One catch worth knowing before you architect around it: pricing doubles above the 200,000-token mark. The 1M context is real, but the long-context tail is not cheap.
This is xAI explicitly courting the enterprise buyer rather than the X power user — and meeting that buyer where their procurement and security reviews already live.
The benchmark story
Grok 4.3 isn't coasting on availability. On the independent Artificial Analysis evaluations it currently sits at #1 on the Omniscience benchmark, where it also posts the lowest hallucination rate among frontier models — a metric that matters far more to a compliance team than another point of math-contest accuracy.
It also tops a cluster of agentic and domain tests:
- #1 on Tau2 Telecom, which scores real-world tool-calling in customer-support agent scenarios
- #1 on Vals AI Case Law, a legal-reasoning benchmark
- #1 on Vals AI Corporate Finance
The throughline is applied reliability: tool use, low hallucination, domain reasoning. That's a deliberate pitch to companies deploying Grok as an agent backbone, not a chatbot.
The infrastructure twist
Here's the genuinely interesting part for engineers. Grok 4.3 on Bedrock runs on Mantle, a new inference engine AWS built into Bedrock and tuned for price-performance, with native support for tool calling, structured output, and response streaming. Reasoning is always on — configurable through a four-level effort parameter (none, low, medium, high) rather than a hard switch.
Coverage of the launch notes Mantle exposes an OpenAI-compatible interface, which means teams already calling OpenAI-style endpoints can point at Grok 4.3 with minimal rewiring. The broader signal: AWS is standardizing model access around familiar wire formats, regardless of which lab built the model.
Why this matters
For most of its life, Grok was a walled garden: great if you lived inside X, awkward if you didn't. Putting Grok 4.3 on Bedrock erases the single biggest objection enterprise buyers had — "we'd have to onboard a whole new vendor." Now it's a checkbox inside an AWS console they already audit.
It also intensifies the price war on Bedrock specifically. With Grok 4.3 at $1.25 / $2.50 and topping a credible hallucination benchmark, the "good enough and cheaper" tier just got a serious new entrant — and the labs already parked on Bedrock now have a reason to sharpen their own pricing.
The 200K pricing cliff is the asterisk. If your workload genuinely needs the full million-token window on every call, model the cost carefully before committing — the sticker price and the real bill can diverge fast.
The Bottom Line
Grok 4.3 on Bedrock is less about a new model and more about distribution: xAI finally going where enterprise budgets already are, with competitive pricing and a benchmark résumé built around reliability instead of bravado. Watch whether xAI keeps publishing on third-party clouds — if Grok keeps showing up wherever buyers already are, the "X-only" era is officially over.
