# OpenAI Flagships Land on Bedrock at First-Party Prices — No Markup

Times of AI Desk · 2026-04-28 · Products

[https://timesof.ai/2026/04/aws-openai-models-bedrock-gpt-5-5-codex](https://timesof.ai/2026/04/aws-openai-models-bedrock-gpt-5-5-codex)

> AWS puts GPT-5.5, GPT-5.4, and Codex on Amazon Bedrock with OpenAI-matching rates, no add-on fees, and usage counting toward AWS commitments. The trade is multi-cloud OpenAI inside existing IAM/PrivateLink stacks — vendor performance claims on Bedrock’s inference engine remain unaudited.

Enterprises that already bought Bedrock governance rarely want a second security model for OpenAI. AWS’s pitch: **frontier OpenAI at list parity inside the controls you already run** — not a new island of API keys.

**AWS** (April 28) announced **GPT-5.5**, **GPT-5.4**, and **Codex** on **Amazon Bedrock**. Access via Bedrock’s Responses API and orchestration tools; pricing matches OpenAI first-party rates with **no additional fees**; usage counts toward existing AWS cloud commitments; IAM, PrivateLink, guardrails, encryption, CloudTrail, and compliance frameworks apply. Models run on Bedrock’s next-generation inference engine (AWS claim).

## What shipped

| Surface | Detail (AWS / About Amazon primary) |
|---------|--------------------------------------|
| **Models** | GPT-5.5, GPT-5.4 alongside Anthropic, Meta, Mistral, Cohere, others |
| **Pricing** | Matches OpenAI first-party; no Bedrock markup stated |
| **Billing** | Counts toward AWS commitments |
| **Codex** | Bedrock API, Codex CLI, desktop app, VS Code extension; inference via Bedrock + AWS credentials |
| **Managed agents** | OpenAI agent harness + Bedrock infra; per-agent identity, action logging; complements AgentCore |

AWS frames Codex as used by “millions weekly” for automation, refactoring, testing — **OpenAI/AWS marketing figure**, not an independent telemetry audit.

## Claims vs checks

Availability, price parity, and control inheritance are **AWS primary**. “High performance, reliability, and security” on the next-gen inference engine is **vendor**. No independent latency or quality delta vs OpenAI first-party is in the announcement. Multi-model Bedrock choice is real product surface; “for the first time” frontier OpenAI on Bedrock is AWS’s narrative of this launch.

## Limits

- Price match is stated; region-by-region SKU matrices are not fully reproduced in the press piece.
- Managed-agent quality vs DIY AgentCore is aspirational until customer case studies ship.
- “Millions weekly” Codex users is un-audited marketing.

## Sources

- [Amazon: “OpenAI models GPT-5.5 and GPT-5.4—and Codex—now on Amazon Bedrock”](https://www.aboutamazon.com/news/aws/bedrock-openai-models) (April 28, 2026).
- AWS announcements on Bedrock integration, Managed Agents, and expanded partnership details.
- Related coverage confirming availability and pricing parity.