OpenAI’s GPT-5.5, GPT-5.4 and the Codex coding agent are now generally available on Amazon Bedrock, giving developers hosted inference access to the company’s latest large language models and a code-focused agent. The move opens Bedrock as a direct deployment channel for OpenAI models and is intended to let organizations run high-performance model inference inside their chosen Bedrock region, which matters for latency, data residency and enterprise compliance.
OpenAI positions GPT-5.5 for the most demanding customer workloads and GPT-5.4 as the better price‑performance option; both are promoted as well suited for coding, complex reasoning, agentic workflows and other professional tasks. Bedrock serves these models through the OpenAI Responses API running on a next-generation inference engine designed for high performance and reliability, exposing features like model — managed multi — turn state and hosted tools.
Codex, described as OpenAI’s coding agent, is available on Bedrock with GPT-5.5 powering inference. OpenAI says more than 4 million developers use Codex weekly to write, refactor, debug, test and validate code across large codebases. The Bedrock — hosted Codex package includes a dedicated desktop application, a command — line interface and IDE integrations for Visual Studio Code, JetBrains and Xcode, with model calls routed through the Responses API.
AWS provides concrete examples for calling the Bedrock — hosted OpenAI models: developers can point the OpenAI SDK or curl at a Bedrock 'mantle' endpoint (a region — specific URL) and set the model ID to openai.gpt-5.5 or openai.gpt-5.4. Sample workflows show using the OPENAI_BASE_URL and OPENAI_API_KEY environment variables and demonstrate Responses API capabilities such as model — managed multi — turn state, hosted tools and function/tool orchestration to integrate model outputs into applications and agent pipelines.
Data residency and billing are emphasized in the Bedrock offering: all model processing remains within the Bedrock region selected by the customer, and pricing is charged per token with no seat licenses or per-developer commitments. Bedrock’s inference service is described as including built — in security and governance controls intended to help meet enterprise compliance needs and to centralize administrative policy for model usage.
Developer setup details for Codex on Bedrock include two authentication pathways: an Amazon Bedrock API key or the AWS SDK credential chain. If AWS_BEARER_TOKEN_BEDROCK is set in the environment, Codex prefers that bearer token; otherwise it falls back to the SDK credential chain. For Bedrock API-key authentication users are instructed to set model IDs (for example, openai.gpt-5.5, openai.gpt-5.4, openai.gpt-oss-120b or openai.gpt-oss-20b) in ~/.codex/config.toml.
Sources
Replies (0)
No replies in this topic yet.
