OpenAI Models Go Live on Amazon Bedrock

GPT-5.5 is live on Amazon Bedrock, running multi-step tasks autonomously.

The team reports that GPT-5.5, along with GPT-5.4 and Codex, is generally available on Bedrock, a month after Amazon and OpenAI expanded their partnership. The OpenAI models run on Bedrock’s next generation inference engine and are priced to mirror OpenAI’s first party rates, with Codex on Bedrock priced per token and all usage counting toward existing AWS commitments. Bedrock acts as the platform for building and running AI applications and agents at production scale, and it promises high performance, reliability, and security for enterprise deployments.

The most capable OpenAI model on Bedrock, GPT-5.5, grips intent faster and handles long, multi-step tasks autonomously. The release notes describe GPT-5.5 as excelling at writing and debugging code across large code bases, analyzing data, generating documents and spreadsheets, and operating software across multiple tools until a task is complete. In short, the improvements are most significant in agentic coding and knowledge work, where progress depends on sustaining context and taking action over time. Benchmarks indicate that GPT-5.5 and GPT-5.4 are built for complexity and duration, with capabilities designed to move beyond single-shot prompts toward sustained, intelligent automation.

From an engineering perspective, the Bedrock integration matters as much as the models themselves. Inference runs through Bedrock, and the service provides an isolated queue with automated capacity management, which is meant to reduce latency variability and make production workloads more predictable. The platform is designed so developers can call the models through the Responses API and pay the same per-token rate as direct from OpenAI, without additional fees. That parity matters for teams budgeting AI projects inside AWS ecosystems, since usage can be tied to existing commitments rather than standalone cost centers.

The announcement stresses a few practical implications for teams building AI-powered software agents. The availability of GPT-5.5, GPT-5.4, and Codex on Bedrock expands the toolkit for production-grade coding assistants and knowledge workers who rely on long-running tasks that span multiple tools and data sources. The ability to sustain context over time and take actions across tools is a differentiator when moving from experimental prompts to production workflows, and the team notes that the improvements are especially visible in agentic coding and knowledge work.

Two broad implications emerge for practitioners. First, cost and governance become more predictable for Bedrock users; pricing alignment with OpenAI first-party rates and AWS commitments help shape budgeting and multi-cloud planning. Second, reliability and observability become more critical as tasks scale. While Bedrock’s isolated queue and capacity management provide a foundation for stable production runs, operators should still design robust monitoring around long-running, multi-tool tasks and guardrails to manage drift, tool failures, or unintended action sequences.

Parameter counts were not disclosed in the release, so teams cannot compare model sizes at a glance. That said, the practical takeaway for developers is clear: OpenAI models on Bedrock are now production-ready within the AWS ecosystem, offering advanced coding and knowledge-work capabilities at familiar pricing, with a stack engineered for reliability in real-world applications.

As the field moves from demo to deployment, the focus for builders will be on crafting robust multi-tool workflows, managing token-based costs, and implementing governance that aligns with enterprise risk profiles. The Bedrock rollout signals a notable push toward integrating frontier models into production software at scale.