How to use OpenAI models on AWS Bedrock
OpenAI GPT models now run on Amazon Bedrock. I enabled them in the console, fixed an AccessDeniedException, and called GPT-6.1 Terra from the playground, the AWS CLI and the OpenAI SDK.
The main idea
You can now use OpenAI models on AWS. In August 2025, the open-source GPT OSS model became available on AWS. On June 1st 2026, OpenAI GPT 5.5 and 5.4 became available. In the Bedrock console, open the OpenAI section to see which models you have access to. That can depend on what your administrator allows.
In the playground I hit an error. I found that I had put a restriction on my own account to block AI model calls, after a very large AWS bill when my agents made lots of model calls. I removed the policy for the demo and got a haiku from GPT-6.1 Terra.
The playground can export JSON and show the API request as an AWS CLI command, and I ran a story prompt that way. For an application, use the OpenAI SDK, set the provider to Bedrock and pass the Bedrock bearer token.
In compare mode I ran Terra against Sol, which is the bigger and more expensive model, on explaining Kubernetes in simple terms. The faster model took about 3 seconds against 7.4 seconds. The model catalog shows 152 models.
Key points
- GPT OSS reached AWS in August 2025. GPT 5.5 and 5.4 reached AWS on June 1st 2026.
- Check access in the Bedrock console under OpenAI. Your administrator may control it.
- An AccessDeniedException in my case came from a policy I had set to block model calls.
- The playground shows the API request as an AWS CLI command you can copy.
- With the OpenAI SDK, change the provider to Bedrock and pass the Bedrock bearer token.
- Compare mode ran Terra against the bigger and more expensive Sol, and the faster one took about 3 seconds against 7.4.
Chapters
- 0:29Finding OpenAI models in Bedrock
- 1:20The error and the fix
- 2:23Calling from the AWS CLI
- 3:09Using the OpenAI SDK
- 3:56Compare mode: Terra and Sol
Save my free seat →