OpenAI Just Went GA on a Rival's Cloud — GPT-5.5 and Codex Are Now Generally Available on AWS Bedrock

OpenAI started selling, officially, on a cloud that isn't OpenAI's

On June 1, a slightly strange scene played out. OpenAI's frontier models GPT-5.5 and GPT-5.4, and the coding agent Codex, reached general availability (GA) on Amazon Bedrock. OpenAI has long meant Microsoft Azure. Now that same OpenAI is being sold — formally — on rival Amazon's cloud.

What launched as a limited preview in April moved to full GA about a month after the two companies announced an expanded strategic partnership. Millions of AWS customers can now call OpenAI models and Codex directly inside the platform they already use. GPT-5.5 is available in US East (Ohio); GPT-5.4 in US East (Ohio) and US West (Oregon). Pricing matches OpenAI's first-party rates — you call them via Bedrock's Responses API and pay the same per-token rate with no additional fees.

Why does this matter? It shows AI models shifting from "a proprietary product locked to one cloud" to "a standard component you call from any cloud." Models become available everywhere, and the real differentiation moves to "where and how you run them cheaply and reliably."

The players — OpenAI, AWS, and the delicate relationship between them

OpenAI has been deeply tied to Microsoft — Azure as primary infrastructure, MS as largest investor. That cemented the perception of "OpenAI models = Azure only." But for OpenAI, depending on a single cloud is risky: it shrinks leverage and narrows customer reach. Hence the pivot to a multi-cloud strategy, pushing its models onto multiple clouds.

AWS is the #1 cloud by share, but early in the generative-AI wave its weakness was "no star model like OpenAI." It has its own models (Nova) and a big investment in Anthropic (Claude), but the model the market most wanted — OpenAI — was missing. This GA fills that gap. AWS's weapon is a neutral, multi-model platform: on Bedrock you pick Claude, Nova, and now OpenAI.

Their relationship is delicate because AWS and MS are head-to-head cloud rivals. OpenAI being both an MS ally and a tenant on MS's competitor cloud is also a signal: "we can do this without MS." Overlay that with yesterday's Build, where MS laid a "Windows agent platform," and you see OpenAI and MS doing a careful dance — interdependent, yet each building independence.

What actually went GA — and what you can do with it

Here are the key facts in a table.

Item	Detail	Note
GA date	June 1, 2026	April limited preview → full GA
Models	GPT-5.5, GPT-5.4	Frontier models
Agent	Codex	OpenAI coding agent
Regions	GPT-5.5: US East (Ohio) / GPT-5.4: East (Ohio) & West (Oregon)	Expansion expected
Pricing	Same as OpenAI first-party, no extra fees	Responses API
Partnership	Expanded strategic partnership ~a month earlier	OpenAI-AWS

Per OpenAI, GPT-5.5 and GPT-5.4 are strong at coding, reasoning, agentic workflows, and complex professional work — use GPT-5.5 for the hardest customer workloads and GPT-5.4 for best price-performance. Codex is OpenAI's coding agent; with GPT-5.5 powering inference, it introduces a class of intelligence optimized for complex, long-horizon developer work.

Codex can be accessed via the Codex App, the Codex CLI, and IDE integrations including VS Code, JetBrains, and Xcode. In other words, developers can summon OpenAI's coding agent from the tools they already use. For AWS, that's a full package: reason with OpenAI models inside Bedrock, then code with Codex.

Who gains — OpenAI, AWS, and enterprise customers

For OpenAI, this is distribution expansion and MS-dependence diversification. It opens a path to push its models at the millions of customers of the world's #1 cloud, while reducing the risk of being tied to one cloud (Azure). Matching first-party pricing is shrewd — the message "you lose nothing using us on Bedrock" pulls AWS customers in.

For AWS, the last piece of the lineup snaps into place. It patches the "no most-wanted model" weakness and completes its neutral-hub position: "come to Bedrock and choose Claude, Nova, OpenAI in one place." If MS targeted the client side yesterday with a Windows agent platform, AWS just hardened its server-side model-distribution hub.

For enterprise customers, choice and leverage grow. If your data and infra already live on AWS, you can use OpenAI models in the same environment without exporting data. Comparing and switching across multiple models via one API reduces lock-in and makes it easier to optimize cost/performance per workload. The choice of "where to call the model from" moves into the customer's hands.

Historical parallels — when a "proprietary part" becomes a "commodity part"

A core component once locked to one platform getting released everywhere is a recurring tech pattern — with both wins and loss of control.

Success — Intel CPUs across many OEMs. Intel didn't supply one PC maker exclusively; it shipped chips to many OEMs and became the market standard. "Intel everywhere" delivered both share and leverage. OpenAI pushing models beyond Azure onto AWS and other clouds rhymes with that — "OpenAI everywhere" lets it chase the standard position without being at the mercy of one cloud's policies.

Success — Netflix's multi-device strategy. Netflix refused to be trapped on one device, putting its app on every platform to maximize reach. Models are similar — the more clouds and dev tools they live on, the bigger OpenAI's usage and influence. Wiring Codex into VS Code, JetBrains, and Xcode extends that same "everywhere" play.

Loss-of-control risk — outsourcing distribution. Conversely, leaning on platforms for distribution can hand the customer relationship and data to the platform. Just as app makers got squeezed by app-store fees and policies, OpenAI spreading across AWS, Azure, and Google Cloud grows reach but can blur whether "the customer uses OpenAI directly, or through a cloud." Multi-cloud is a tightrope between reach and control.

Competitor counter-plays — Anthropic, Google, Microsoft

Anthropic (Claude) has been AWS's longstanding key partner. The de facto flagship frontier model on Bedrock was effectively Claude — and now OpenAI sits on the same shelf. For Anthropic, being directly compared inside AWS creates pressure to sharpen its differentiation on coding, agents, and safety. With Anthropic reportedly in an IPO process (separate coverage), this rivalry will draw even more attention.

Google counters with Gemini and Vertex AI, its card being a vertically integrated "own model + own cloud." If OpenAI grows reach by riding others' clouds, Google can push price/performance with the efficiency of owning TPUs, Gemini, and Vertex under one roof. Multi-cloud neutrality vs. vertical-integration efficiency — that's the matchup.

Microsoft holds the most complicated position. As largest investor and Azure provider, it has to watch its partner formally land on a rival cloud. But MS grabbed the client OS with yesterday's Windows agent platform and is lowering dependence with its own MAI models. So as OpenAI goes multi-cloud, MS draws its own "we're fine without OpenAI" picture — both building independence.

So what actually changes — by persona

If you're a developer or enterprise on AWS, you can now use OpenAI's latest models without switching clouds. Compare Claude, Nova, and OpenAI through one Bedrock API and pick the cheapest, best-fit model per workload. In regulated settings where data can't leave AWS, this is a genuinely large advantage.

If your team is weighing coding automation, the key is Codex landing in VS Code, JetBrains, and Xcode. You can call OpenAI's coding agent right inside your usual IDE, so it's worth piloting long-horizon dev work delegated to an agent. Just note pricing matches OpenAI first-party, so estimate costs for token-heavy agent workflows in advance.

If you're a solo developer or founder, look at the bigger picture. When a model becomes "a commodity part callable from any cloud," real edge shifts from "owning a model" to "what you build on top of it." In an era where everyone uses the same model at the same price, differentiation comes from product, domain, and user experience. The more model access is leveled, the more opportunity actually opens for solo builders with ideas and execution.

References

• AWS Machine Learning Blog — OpenAI models and Codex on Amazon Bedrock now GA
• Get started with OpenAI GPT-5.5, GPT-5.4 models, and Codex on Amazon Bedrock (AWS Blog)
• GPT-5.5, GPT-5.4, and Codex now generally available on Amazon Bedrock (AWS What's New)
• OpenAI models GPT-5.5 and GPT-5.4 — and Codex — now on Bedrock (About Amazon)
• OpenAI — Frontier models and Codex are now available on AWS