Amazon Web Services sells OpenAI’s GPT-6.1 Terra on Amazon Bedrock at $1 per million input tokens and $5 per million output tokens through global routing, but keeping requests inside the US costs $1.10 and $5.50, according to Amazon’s Bedrock model card. The card says “Commercial In-Region prices include a 10% fee over OpenAI rates”.
By our count of the same card’s region table, 25 of the 31 AWS Regions where the card marks the model as available reach it only through global cross-Region inference, which AWS describes as routing “anywhere worldwide when there are no residency constraints”. All eight Regions the card lists in Europe, from Frankfurt and Zurich to London and Paris, are among them.
A short Business Engineer explainer on the difference between an AI model and the harness around it. It teaches the general idea only and says nothing about any company in this piece.
The key insight: By our count of Amazon’s model card, 25 of the 31 AWS Regions offering GPT-6.1 Terra reach it only through global routing, and keeping requests in the US costs 10% more, the same premium OpenAI lists for regional processing on its own API.
AWS announced general availability on 26 October, the model launch date on its card, and made it the lead item of its Weekly Roundup on 2 November. For prompts of 272K input tokens or fewer, the card lists $1.10 per million input tokens, $0.055 for cache reads and $5.50 for output, both in-Region and through geographic cross-Region inference. Global cross-Region inference costs $1.00, $0.05 and $5.00.
Above 272K input tokens the prices rise to $2.20 input and $8.25 output in-Region or by geography, and $2.00 and $7.50 globally. The card gives a context window of 1,050,000 tokens and a maximum output of 128,000 tokens.
Output price per 1 million tokens for prompts of 272K input tokens or fewer, Standard tier. Bedrock prices from Amazon’s model cards for GPT-6.1 Terra and GPT-5.6 Terra; OpenAI API prices from OpenAI’s gpt-6.1-terra model page. The OpenAI regional-processing ($5.50) and Batch/Flex ($2.50) figures are our arithmetic from OpenAI’s stated 10% premium and 50% discount.
The Same Price as OpenAI’s Own API
OpenAI’s API model page lists GPT-6.1 Terra at $1.00 per million input tokens, $0.05 cached and $5.00 output, and says “Regional processing adds a 10% premium where available”. As we read the two price lists, Bedrock’s global price equals OpenAI’s standard rate, and the 10% in-Region fee mirrors the premium OpenAI charges for regional processing on its own API.
The difference is in the service tiers. Amazon’s card lists Standard as the only supported tier for this model, with Priority, Flex and Reserved marked as not supported. OpenAI’s page says “Batch and Flex prices are 50% lower than Standard”, which puts them at $0.50 input and $2.50 output, by our arithmetic.
Where the Model Runs
On Amazon’s bedrock-runtime endpoint the card lists 32 Regions. Six offer the geographic profile: N. Virginia, Ohio, N. California and Oregon in the US, plus Canada Central and Calgary, whose requests the card routes to Canadian and US Regions. Bahrain is listed with no option marked. The remaining 25, from Frankfurt and London to Tokyo, Mumbai and São Paulo, are marked for global routing only.
Single-Region processing is marked only on a second endpoint, bedrock-mantle, and only in N. Virginia. The two GovCloud Regions appear in that table with no option marked.
OpenAI’s Apps on Amazon’s Inference
AWS says customers can call the model through Bedrock APIs “or configure ChatGPT Work and Codex to use the model on Amazon Bedrock”. In the roundup, AWS’s Daniel Abib wrote that the model was “priced well below the model beside it”, and that OpenAI’s benchmark results are “OpenAI’s figures, not an independent test”.
On Bedrock’s own price tables, GPT-5.6 Terra costs $2.20 input and $13.20 output in-Region, according to its model card. GPT-6.1 Terra’s $5.50 output price is 58% lower, by our arithmetic, and its input price is 50% lower; AWS says both are at least 50% lower. OpenAI says GPT-6.1 Terra scored 15 percentage points higher than GPT-5.6 Terra on its Agents’ Last Exam evaluation.
As we read it, the business lever is distribution: Amazon resells OpenAI’s newest Terra model at OpenAI’s own rates, so for a buyer on AWS the choice is about routing and residency rather than a cheaper or dearer channel.
The Structural Read
As we read it, AWS is not undercutting OpenAI on this model: its global rate equals OpenAI’s standard rate, and its in-Region rate carries the same 10% that OpenAI lists for regional processing.
As we read it, what Bedrock adds is placement inside AWS accounts; AWS says established AWS controls help customers secure workloads, govern access and audit model invocation activity.
As we read it, a lower-priced path for this model sits on OpenAI’s own API, where OpenAI says Flex costs 50% less than Standard, a tier Amazon’s card marks as not supported for GPT-6.1 Terra on Bedrock.
Daniel Abib, AWS Weekly Roundup, 2 November 2026
“what interests me is where it sits rather than what it tops”
Three Implications
FOR AWS BUYERS Global routing at $1 input and $5 output matches OpenAI’s list price; the card marks four US and two Canadian Regions for geographic routing at $1.10 and $5.50.
FOR EUROPEAN BUYERS Amazon’s card marks all eight Regions it lists in Europe for global routing only, which AWS describes as routing anywhere worldwide when there are no residency constraints.
FOR OPENAI AWS says ChatGPT Work and Codex can be configured to use the model on Amazon Bedrock; our compute tracker lists OpenAI’s $250 billion Azure commitment, from Microsoft’s 10-Q.
The framework describes the token this way: “It is simultaneously the unit of cognition the model produces, the unit of compute the data center serves, the unit of price the lab charges, and the unit of value the enterprise extracts.”
As we read it, GPT-6.1 Terra on Bedrock shows the unit of price set by the lab and carried through by the cloud: AWS lists OpenAI’s rate and adds the same 10% for in-Region processing. The framework is the owner’s argument and does not discuss this launch.
The AWS and OpenAI documents we read do not give the revenue split between AWS and OpenAI, usage of OpenAI models on Bedrock, or any timetable for European geographic routing for this model. The benchmark results are OpenAI’s own. The 25-of-31 count, the 58% figure and the Batch and Flex prices are our arithmetic from the published tables. We did not contact AWS or OpenAI.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
GPT-6.1 Terra on Amazon Bedrock costs $1 per million input tokens and $5 per million output tokens through global routing, the same as OpenAI’s standard API price, and 10% more in-Region or through geographic routing, according to Amazon’s model card.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Prices, service tiers and Region availability come from Amazon’s Bedrock model cards for GPT-6.1 Terra and GPT-5.6 Terra, read on 2 November; OpenAI’s prices come from its API model page. The 25-of-31 Region count, the 58% figure and the Batch, Flex and regional-processing prices are our arithmetic. We did not contact AWS or OpenAI.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
