OpenAI models now on Amazon Bedrock
- Amazon Web Services said on June 1 that OpenAI's GPT-5.5, GPT-5.4 and Codex are now generally available on Amazon Bedrock. - Amazon said pricing matches OpenAI's first-party rates, while OpenAI said Codex is used by more than 5 million people weekly. - Amazon Bedrock documentation and OpenAI developer guides now outline setup, supported features, and AWS-managed controls for enterprise deployments.
Amazon Web Services said on June 1 that OpenAI's GPT-5.5, GPT-5.4 and Codex are now generally available on Amazon Bedrock, extending OpenAI's newest models and coding agent into AWS's managed AI platform. OpenAI said the rollout gives customers two paths inside AWS: Bedrock access to its frontier models and Codex for software engineering work. The launch was announced a month after AWS and OpenAI previewed a broader partnership around Bedrock-managed agents and OpenAI-powered development tools. ### Which OpenAI products are actually included on Bedrock? AWS said the generally available lineup includes GPT-5.5, GPT-5.4 and Codex. In its launch note, Amazon described GPT-5.5 as OpenAI's "most capable" model for agentic coding, data analysis and multi-step tasks, while AWS documentation says GPT-5.4 supports reasoning, coding, computer use, long-context workflows and tool use. (aws.amazon.com) OpenAI said Codex on Amazon Bedrock lets teams write, review, debug and modernize code inside AWS environments they already use. The company said Codex is used by more than 5 million people every week. ### How does this work differently from using OpenAI directly? Amazon said customers can use the models through Bedrock with the same AWS security, governance and operational controls they already use across the cloud platform. (aws.amazon.com) OpenAI's developer documentation says Bedrock availability is aimed at organizations that want procurement, identity, regional controls and related cloud operations to stay inside AWS-managed infrastructure. (openai.com) OpenAI said Codex can also be configured to use Amazon Bedrock as the model provider, with requests sent through AWS-managed authentication and access controls. Its documentation says the OpenAI-hosted Responses API is not in the request path in that setup. ### What is AWS emphasizing for enterprise buyers? Amazon said pricing for GPT-5.5 and GPT-5.4 on Bedrock matches OpenAI's first-party rates and carries no additional fees. (aws.amazon.com) Amazon also said usage counts toward AWS cloud commitments, a detail the company highlighted again in its public news post about the launch. AWS said Bedrock customers can use OpenAI models alongside its broader agent stack, including Bedrock AgentCore and Bedrock Managed Agents powered by OpenAI. (developers.openai.com) In April, Amazon said those managed agents combine OpenAI frontier models with the OpenAI agent harness for long-running tasks on AWS. ### Why is Codex part of the announcement? OpenAI said Codex on Bedrock is meant for AI-powered software development, not just chat or text generation. (aboutamazon.com) Amazon's June 1 blog post framed the launch around production workloads and software development, while OpenAI's Bedrock materials focus on coding, debugging and modernization tasks. (aws.amazon.com) AWS also said Codex is available through the Codex app, command-line interface and IDE integrations, with inference routed through Amazon Bedrock and protected by AWS controls including IAM, VPC isolation and encryption. ### What security questions are still hanging over frontier models? Cisco said in a research post published in late May that no tested frontier model was immune to multi-turn attacks. (aws.amazon.com) The company said its AI Threat Intelligence and Security Research team evaluated 15 closed models from OpenAI, Anthropic, Google, Amazon and xAI and found materially higher attack success rates in multi-turn settings than in single-turn tests. (aboutamazon.com) Cybersecurity Dive, citing Cisco's findings, reported that multi-turn malicious prompt success rates ranged from 8% to 88%, versus 2% to 65% for single-turn prompts. Cisco researchers DJ Conley and Chawin Chang wrote that relying on published single-turn scores for business decisions presents "security and governance risk," according to that report. (blogs.cisco.com) ### What happens next for customers that want to use it? Amazon's Bedrock product page now lists OpenAI models as generally available, while AWS and OpenAI have both published setup guides for Bedrock deployments. OpenAI also maintains a sign-up page for customers seeking access to OpenAI models, Codex and Bedrock-managed agents on AWS. (aws.amazon.com) (cybersecuritydive.com)