OpenAI's GPT-5.5, GPT-5.4, and Codex hit general availability on Amazon Bedrock

OpenAI's two frontier chat models and its Codex coding agent are now generally available on Amazon Bedrock. AWS announced the move on June 1, 2026, one month after the two companies expanded a partnership that first brought OpenAI inference inside AWS at the end of April. Customers can now call GPT-5.5, GPT-5.4, and Codex through Bedrock's Responses API, with pricing matched to OpenAI's first-party rates and usage that counts toward existing AWS commitments.

What's new

• Three OpenAI products move from limited preview to GA on Bedrock: GPT-5.5 (positioned as OpenAI's most capable model), GPT-5.4 (positioned for best price-performance), and Codex (the OpenAI coding agent).
• Bedrock model IDs are openai.gpt-5.5 and openai.gpt-5.4, called through the Responses API on Bedrock's next-generation inference engine. The previously released openai.gpt-oss-120b and openai.gpt-oss-20b open-weight models also remain on Bedrock.
• Pricing matches OpenAI first-party rates with no Bedrock surcharge. Codex carries pay-per-token pricing with no seat licenses and no per-developer commitments.
• Enterprise-grade controls come included: IAM permissions, VPC and PrivateLink isolation, KMS encryption, and AWS CloudTrail audit logging. AWS states that prompts and responses are not used to train models and are not shared with model providers.
• For customers with data residency requirements, processing stays in the Bedrock Region the customer selects. Region availability varies per AWS documentation.
• Codex is accessible via the Codex app, CLI, and IDE integrations, with all inference routed through Bedrock.

Context

The OpenAI-on-AWS arrangement was announced April 28, 2026 alongside an expanded multi-year compute commitment, and the open-weight gpt-oss models landed on Bedrock first. Today's GA brings the closed frontier models into the same surface. It tracks a broader pattern: hyperscalers are no longer single-lab platforms. AWS now offers OpenAI's frontier alongside Anthropic's, which has had its own Claude Platform on AWS since May 11, 2026 (with webhooks, multi-agent orchestration, and self-hosted sandboxes added on May 29). Customers can pick OpenAI or Anthropic — or both — without leaving the same IAM and VPC perimeter.

The Codex piece is the more pointed strategic move. AWS quotes OpenAI's figure of more than 4 million developers using Codex each week, and pulling that audience inside Bedrock puts AWS in direct competition with GitHub-on-Azure for the agentic-coding workflow.

Why it matters

This closes a real gap. Enterprises that picked AWS as their cloud and OpenAI as their model previously had to stitch them together through OpenAI's own API plus AWS billing — outside the unified IAM and audit story most large customers want. With Bedrock as the routing layer, prompts inherit existing AWS controls and usage counts against existing AWS spend commitments, which is the procurement detail that actually moves enterprise adoption. It also formalizes a multi-frontier reality on AWS: customers who want OpenAI and Anthropic side-by-side on the same provider get that today, with the same security model around both. For Bedrock, it removes the longstanding objection that the platform shipped every frontier model except OpenAI's; for OpenAI, it broadens distribution into accounts where Azure isn't the default; for AWS coding-tool strategy, putting Codex inside the Bedrock perimeter is a direct response to the GitHub-Copilot incumbency.