OpenAI Brings GPT-5.5 and Codex to General Availability on Amazon Bedrock and Challenges Azure's Monopoly
GPT-5.5 at $30 and $180 per million tokens, the same rate as OpenAI direct. Bedrock becomes a one-stop shop where Claude, GPT, Llama, and Mistral compete for corporate POCs.
On Monday (1st), AWS made models GPT-5.5, GPT-5.4, and the Codex code agent generally available on Amazon Bedrock. The offering includes the Bedrock Managed Agents service powered by OpenAI, which delivers custom managed agents in an AWS environment. This move concludes a strategic partnership cycle that began in April when the models entered limited preview and was accelerated to GA in less than two months due to corporate demand pressure, according to Amazon.
Pricing is the most striking aspect. GPT-5.5 Pro costs $30 per million input tokens and $180 per million output tokens, exactly the same rates charged by OpenAI for direct access. AWS does not embed a markup. For short-context GPT-5.5, the tariff drops to $12.50 and $75. The client pays for consumed tokens, with no seat licence and without committing to a minimum subscription per developer, and the expenditure counts against corporate commitments made with Amazon, such as the Enterprise Discount Programs.
The Corporate Barrier Has Fallen
The migration of OpenAI's models to Bedrock addresses one of the most common objections raised by CIOs in vendor committees since 2024. The versions hosted on Bedrock inherit the entire set of controls that AWS customers already operate: IAM for access management, AWS PrivateLink for isolated connectivity, configurable guardrails, encryption at rest and in transit, and the same contractual clause of not training on customer data. Compliance that previously required a parallel vendor approval process now fits within the existing Bedrock, with the same DPA, BAA contracts and data residency clauses.
Amgen and Autodesk were introduced as the first adopters of the GA phase. Other companies were in preview. Corporate demand, according to AWS, justified the leap to general availability at an aggressive pace even by Bedrock standards, which typically keep models in preview for six months or more before broad release.
The Competitive Geometry Among Hyperscalers
AWS arrives at OpenAI's models later than Azure, which has been operating the Azure OpenAI Service since 2023 based on the Microsoft-OpenAI commercial partnership established in 2019. The practical difference is twofold. First, Azure still holds contractual precedence over frontier versions before AWS, according to executives from both companies. Second, AWS offers Bedrock as a one-stop shop: customers can compare GPT-5.5 with Claude Opus 4.7, Meta's Llama 3, Mistral Large, and Cohere's Command R+ in the same console and switch models without redoing integration.
For Anthropic, which has AWS as its preferred computing provider, OpenAI's arrival at the same counter is a test. Claude had Bedrock as an almost exclusive showcase in the enterprise segment. Now it shares the stage with a direct competitor, under the same SLA and the same customer procurement committee. The timing is particularly uncomfortable: Anthropic filed a confidential IPO on the same date and needs to demonstrate corporate revenue growth in the upcoming quarters.
How This Plays Out Outside the US
Regional availability is selective. GPT-5.5 runs only in the US East (Ohio); GPT-5.4 in both US East (Ohio) and US West (Oregon). European customers need to either move data or accept transatlantic latency, which clashes with post-Schrems II expectations in jurisdictions like Germany and France. AWS has signalled an expansion to Frankfurt and Dublin "in the coming months," without a firm date. European regulators in telecom, finance, and health are likely to maintain Azure OpenAI as the preferred path until this gap closes, and Microsoft is aware of this.
In Japan and Singapore, where MUFG and DBS accelerated their generative AI pilot in 2026, price parity with OpenAI eliminates internal friction. MUFG's AI committee was already operating on Bedrock for Claude and Llama; fitting GPT-5.5 into the same governance gateway makes comparative supplier testing almost trivial. The competition for internal POCs is expected to intensify over the next eight weeks, with the winner of the shootout determining an eight-digit annual contract.
The broader movement is more subtle. OpenAI is no longer exclusive to Azure in the enterprise front. AWS forces direct comparison with Anthropic within Bedrock itself. Customers gain in price. The hyperscalers lose the moat of "single model available here." The year 2026 began with the boundaries between the three major cloud providers becoming harder to defend.