xAI’s Grok 4.3 is now available on Amazon Bedrock at $1.25 per million input tokens and $2.50 per million output, with a 131K context window. At those specs it is the cheapest major-lab model on Bedrock, and that alone makes it worth a look for anyone whose 2026 AI budget is already a smoking crater.
The comparison you actually want
Put the output prices side by side. Grok 4.3 at $2.50. Claude Sonnet 5 at $10 introductory (heading to $15). GPT-5.5 at $15. For a large class of workloads — classification, extraction, routing, summarizing, first-draft grunt work — you are not paying for reasoning brilliance, you are paying to move text. On those jobs, a model at a quarter of the price that gets it right is not a compromise, it is the correct answer.
The AI tool stack actually worth paying for
One email a week. The models, tools and moves that matter, stripped of hype and filtered so you don’t have to drink from the firehose. Free, and you can bail anytime.
Where cheap stops being smart
Do not put the budget model on your hardest agentic chains and then act surprised when it wanders off the plan. The whole point of Sonnet 5’s launch was reliability on multi-step tool use — that is a real gap, and it is exactly where a cheap model quietly costs you more in retries, cleanup and blown outputs than you saved on tokens. The move is not “pick the cheapest.” It is tiering: route the boring 80% to Grok 4.3 and reserve the expensive models for the work that actually needs them.
The Bedrock angle
Bedrock now carries Claude, Amazon’s own Nova, Llama, Mistral, and now Grok under one AWS contract and one bill. If you are already an AWS shop, adding Grok 4.3 to your routing layer is a config change, not a procurement saga. That convenience is underrated — and it is how AWS keeps you from wandering off to a rival’s API.