What’s now generally available

The reporting from AWS Machine Learning Blog, OpenAI’s official announcement, AWS News Blog, and Help Net Security confirms:

• Models GA on Amazon Bedrock:
  • GPT-5.5 — OpenAI’s current frontier model
  • GPT-5.4 — previous-generation frontier
  • Codex — OpenAI’s coding agent
• Plus Managed Agents (limited preview from earlier April 28 announcement now expanded)
• GA date: June 1-2, 2026
• Pricing: matches OpenAI first-party rates (no AWS markup)
• AWS commitment integration: usage counts toward existing AWS contracts and commitments
• Inference platform: Bedrock’s next-generation inference engine
• Security: IAM authentication, VPC isolation, encryption — full AWS-native security stack
• Codex distribution: App, CLI, and IDE integrations all available with inference routed through Bedrock

How this came about — the April 27 → June 2 arc

The story makes sense only in the context of the April 27 Microsoft-OpenAI restructuring:

Six weeks ago, “OpenAI runs on Azure exclusively” was a structural fact. Today it’s already wrong. By August, “OpenAI inside Microsoft’s flagship developer product” will also be wrong. The decoupling is happening layer by layer, on a timeline that the April 27 deal effectively set in motion.

Why this matters for enterprise customers

Three reads.

1. The “Azure lock-in” objection to OpenAI is functionally dead. Through 2024-2025, the standard enterprise procurement objection to OpenAI was “we’d have to spin up Azure tenancy.” AWS-shop customers (which is most large enterprises by spend) had structural friction. As of today, that friction is gone — OpenAI runs natively inside AWS environments customers already operate.

2. Usage counting toward AWS commitments is procurement gold. Large enterprise AWS contracts include committed spend tiers (Enterprise Discount Programs, Reserved Instances). Having OpenAI usage count against those commitments means companies can deploy GPT-5.5 without expanding their cloud-vendor footprint or negotiating new contracts. For procurement teams, this is a meaningful workflow simplification.

3. Multi-cloud serving improves reliability and pricing. OpenAI is now serving GPT-5.5 inference across Microsoft Azure (primary), AWS Bedrock (new), and Google Cloud Vertex AI (existing, smaller). Each cloud provides geographic redundancy, capacity-burst overflow, and competitive pricing pressure. For end users, this should translate into more reliable serving and continued downward pressure on per-token costs.

The Daybreak next step

The OpenAI announcement specifically called out Daybreak — OpenAI’s cybersecurity initiative — as the next major capability arriving on AWS Bedrock. Daybreak combines OpenAI’s frontier models with Codex Security for vulnerability discovery and remediation workflows.

The strategic context: Daybreak is OpenAI’s answer to Anthropic’s Project Glasswing. The competitive dynamic in frontier-grade AI cybersecurity tooling is now:

• Anthropic: Claude Mythos via restricted-access Project Glasswing (announced first, with 13+ partner organizations including AWS, Apple, Google, Microsoft)
• OpenAI: GPT-Daybreak + Codex Security, launching on AWS Bedrock

Both labs are pursuing “frontier model finds and fixes vulnerabilities at scale” with different deployment topologies. Anthropic restricts; OpenAI distributes through hyperscaler clouds.

What it means for ChatGPT and Codex users

For ChatGPT consumer subscribers: nothing changes operationally. Your subscription accesses OpenAI’s hosted ChatGPT product directly; Bedrock GA affects only the enterprise API path.

For Codex users: this is the bigger functional change. Codex App + CLI + IDE all now have AWS-native deployment options. For developers in AWS-shop environments (where security teams require AWS-native auth, encryption, VPC isolation), Codex is now usable without architectural workarounds.

For enterprise API consumers: pricing matches first-party rates, usage counts toward AWS commitments. The structural objection to going through Bedrock vs OpenAI direct is now “do I want AWS-native security wrapping the API call” — and for most enterprise AWS shops, the answer is yes.

For Claude Code users: the competitive landscape now has more credible Codex deployment paths. For developers evaluating Claude Code vs Codex, the friction differential narrows. Direct quality + workflow comparison matters more.

What it means structurally for the harness category

For the AI Harnesses category, this is operationally significant:

• OpenAI Codex — now AWS-native, enterprise-deployable without Azure dependency
• GitHub Copilot — still primarily Azure but moving to Microsoft’s own Polaris starting August
• Claude Code — Anthropic’s flagship, AWS-native (Anthropic’s primary cloud relationship is AWS)
• Antigravity — Google-only by design
• Open-source harnesses (OpenCode, Aider, Pi) — multi-model, multi-cloud by definition

For enterprise harness selection, the cloud-deployment story used to be a meaningful differentiator. As of today, OpenAI/Codex and Claude Code both deploy AWS-natively; GitHub Copilot stays Azure-native; Antigravity stays GCP-native. The “where does my code go” question is now substantially less about the harness vendor and more about the enterprise’s existing cloud-procurement posture.

The honest caveats

Three caveats:

Microsoft Azure remains OpenAI’s primary cloud. AWS Bedrock GA is additive, not a replacement. The bulk of OpenAI inference still runs on Azure. AWS Bedrock is a procurement and reliability option, not a strategic re-platforming.

“Pricing matches OpenAI first-party rates” doesn’t mean it stays that way. AWS and OpenAI both retain pricing flexibility. The current price parity could shift if competitive dynamics change — e.g., if Microsoft cuts Azure OpenAI pricing aggressively to retain customers, AWS might match or undercut, and OpenAI’s direct API pricing might fluctuate too.

Daybreak isn’t shipped yet. OpenAI’s announcement called out Daybreak as the next AWS capability. No GA timeline is published. Treat it as “stated future” not “confirmed coming soon.”

What it changes for Pick Right readers tomorrow

If you’re a ChatGPT Plus subscriber, nothing changes. If you’re a Codex user on an AWS-shop team, the AWS-native deployment path is now available — talk to your security team about Bedrock-routed Codex if Azure-routed was previously blocked.

For broader context, see the ChatGPT review, the Codex review, the AI Harnesses category, the OpenAI Frontier Governance Framework news, the Microsoft Project Polaris article, and the OpenAI confidential S-1 filing coverage for the broader cloud-decoupling and IPO-pipeline picture.

Questions or corrections? Email Pick Right. Want the full list? See all news.