Amazon says companies can use Astra through Bedrock APIs and connect it to work tools. The service offers a large context window, but retention and permission settings still need review.
New to this? Read it in simple words
- AWS customers can now use GPT-6 Astra through Amazon Bedrock. Amazon says Astra supports up to one million input tokens.
- Amazon says data from using Astra is not used to train the model. AWS says requests marked as possible abuse may be kept for up to 30 days.
- A strong model in a familiar cloud is easier to start using. But teams must still check data rules, access rights and human approval.
- Token
- A small piece of text, often part of a word. AI use is priced per token.
- API
- A way for developers to use a service directly from their own software.
- Inference
- Using a trained AI model to answer questions or do tasks.
Astra enters an existing cloud system
AWS customers can now use GPT-6 Astra through Amazon Bedrock APIs. They can also configure ChatGPT Work and Codex to use Astra on Bedrock. This lets some companies keep model access near their existing cloud controls.
Amazon says Astra supports up to one million input tokens. A large window can hold long documents, code, or business records. It does not mean every part receives equal attention or that a long answer is always correct.
The model can support agents that use tools and finish several steps. Bedrock adds familiar controls for identity, network access, logs, and costs. Teams still need to set those controls correctly for each real task.
AWS says Astra supports up to one million input tokens. Customers still need to review retention and tool permissions.
Permissions follow the connected account
Amazon also describes enterprise plugins for services such as Workday, Navan, and Avalara. A plugin can use the accounts and permissions that a worker already has. This can make setup quick, but it can also copy access that is too broad.
A good agent should receive the smallest set of rights needed for one job. Reading a report is different from changing payroll or buying travel. Important actions should have a clear human approval step.
Teams should test what happens when a tool fails, a page gives harmful instructions, or a request is unclear. A safe workflow needs a stop point, an action record, and a person who owns the result.
Data promises need exact settings
Amazon says inference data is not used to train the model and does not need to be shared with OpenAI. That can matter to companies with strict data rules. Buyers should confirm which service, region, and account settings support the promise.
AWS documentation says traffic flagged for possible abuse may be kept for up to 30 days. Eligible customers can ask their account team about zero retention. This detail deserves review before sensitive work begins.
Bedrock makes Astra easier to buy and connect. The useful question is not only whether the model is strong. It is whether one complete workflow has the right data limits, permissions, checks, cost controls, and recovery plan.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1Primary source · September 8, 2026Take on your most ambitious work with GPT-6 Astra on Amazon Bedrock AWS
- 2
- 3
We used AWS and Amazon for availability, context size, plugins, and data claims. We used the AWS model card for retention details. We treat privacy and performance statements as provider claims and tell buyers to check their exact region and settings.