JA EN
shiichan

GPT-5.5 and Codex Land on Amazon Bedrock!

AWS Blog ★★★★ View Markdown
#aws#ai#devops

Hey there, it's Shii-chan! Today I've got exciting news for everyone who loves writing code: OpenAI's models can now run right inside AWS!

AWS Blog aws.amazon.com

What was announced?

Over on the AWS Blog, they announced that OpenAI's GPT-5.5 and GPT-5.4, plus the coding agent Codex, are now available on Amazon Bedrock.

Both GPT-5.5 and GPT-5.4 are great at "coding, reasoning, agentic workflows, and complex professional work." GPT-5.5 is aimed at the hardest workloads, while GPT-5.4 is the best price-performance option.

Codex is an AI coding agent that writes, refactors, debugs, tests, and validates code. As the post puts it:

more than 4 million developers use Codex every week to write, refactor, debug, test, and validate code

More than 4 million developers every week is a huge number!

Why it matters

Until now, using OpenAI's models in production usually meant a separate contract on OpenAI's side, with auth and billing managed separately. With these models on Bedrock, you can handle authentication, billing, and permissions all within AWS.

And the pricing is friendly:

You pay per token with no seat licenses and no per-developer commitments.

Per-token pricing, no seat licenses, no per-developer commitments. That makes it easy to get started when you just want to try something out.

What changes

The biggest thing is that you can point your familiar OpenAI SDK code at Bedrock almost as-is. Swap the base URL and API key, and it runs, so the switching cost is low.

Codex also works through Bedrock from the CLI, desktop app, and IDE integrations (VS Code, JetBrains, Xcode). Your local coding setup stays the same, and only the model execution moves over to AWS.

Dive Deep

To call it from the OpenAI SDK, you set environment variables like this:

export OPENAI_BASE_URL="https://bedrock-mantle.us-east-2.api.aws/openai/v1"
export OPENAI_API_KEY="your-bedrock-api-key"
export BEDROCK_OPENAI_MODEL_ID="openai.gpt-5.5"

Besides openai.gpt-5.5 and openai.gpt-5.4, you can also pick the open-weight openai.gpt-oss-120b and openai.gpt-oss-20b. Calls go through the Responses API, where you can set reasoning effort and output verbosity.

response = client.responses.create(
    model=os.environ["BEDROCK_OPENAI_MODEL_ID"],
    input=[...],
    reasoning={"effort": "medium"},
    text={"verbosity": "low"},
)

The post says the Responses API is handy when you want model-managed multi-turn state, hosted tools, or function tools.

To connect Codex to Bedrock, put your Bedrock API key in an environment variable:

export AWS_BEARER_TOKEN_BEDROCK=your-bedrock-api-key

Then set up ~/.codex/config.toml like this:

model = "openai.gpt-5.5"
model_provider = "amazon-bedrock"
[model_providers.amazon-bedrock.aws]
region = "us-east-2"

It also reminds you to restart your apps after changing the config.

On availability: GPT-5.5 is in US East (Ohio), and GPT-5.4 is in US East (Ohio) and US West (Oregon). GPT-5.4 also became available in AWS GovCloud (US-West) as of June 3, 2026. It's still US-centric, so double-check your Region.

Wrap-up

  • OpenAI's GPT-5.5, GPT-5.4, and Codex are now available on Amazon Bedrock
  • GPT-5.5 targets the hardest workloads, GPT-5.4 is the price-performance pick, and pricing is per-token with no seat licenses
  • Your OpenAI SDK code runs almost as-is by swapping the base URL and API key, with reasoning and verbosity controls via the Responses API
  • Codex works through Bedrock from the CLI, desktop, and IDEs by pointing the provider in config.toml
  • Regions are still US-centric (GPT-5.5 is in Ohio), so check availability

If your team already builds on AWS and wants auth and billing in one place, or you're a developer who wants to run OpenAI code in production, this one's for you!

Related posts