Amazon Web Services shipped a redesigned Bedrock console on June 5 that natively supports both Anthropic’s Messages API and OpenAI’s Responses and Chat Completions APIs without any code changes from developers. The strategic subtext is more significant than the UX update: AWS is positioning Bedrock as the place where enterprise AI procurement happens, not the model labs themselves.
The mechanics are straightforward. A developer with existing code written against the Anthropic SDK or the OpenAI SDK can point it at the bedrock-mantle endpoint and keep shipping. No rewrites. No new authentication patterns to learn beyond AWS IAM. The friction of switching is reduced to a single environment variable.
The new model catalog consolidates Claude, GPT, and open-weight models such as Llama and Mistral into one browsable interface with consistent capability flags: context window, modality support, tool-use support, and pricing tier. Previously, comparing models across providers meant stitching together documentation from three separate company websites. Bedrock now absorbs that discovery work. This is the Costco move: a single warehouse where you buy Kirkland (Bedrock-native) or name-brand (Anthropic, OpenAI), but you check out through one register, and that register is AWS.
Project-based workflows extend the lock-in further. Teams organize their inference workloads into named projects that each carry their own model selection, usage dashboards, and evaluation harness. The console generates Python, Node.js, and Java code snippets pre-filled with the project’s model ID, endpoint URL, and API key reference. An engineer copies the snippet and runs it without modification. Each line of generated code tightens the organizational habit of routing AI calls through AWS rather than directly through a lab.
The console also supports routing requests from coding agents, including Claude Code, Cline, Codex, Cursor, and OpenCode, through the Bedrock endpoint. That last detail is not trivial. If a team’s agentic coding infrastructure routes through Bedrock, the commercial relationship with the model lab becomes invisible to the team’s purchasing process. AWS captures the billing relationship.
The economics follow from the architecture. Model labs price their APIs directly. When cloud providers intermediate that relationship, they capture a margin on top while abstracting the lab’s price signal from the buyer. The lab gets distribution and AWS gets the enterprise account. That arrangement holds so long as AWS continues adding models fast enough to stay ahead of a developer’s incentive to go direct.
The lock-in battle in enterprise AI is not at the model layer. GPT-4o, Claude Sonnet, and Llama can all be swapped today with modest engineering effort. The harder migration is from a console where your team’s projects, dashboards, evaluation runs, API keys, and IAM roles all live. Microsoft is running the same play through Azure AI Foundry. The competition is not about which model wins; it is about which cloud provider becomes the default procurement surface.
The new console is available across 14 AWS regions as of June 5, covering North America, Europe, Asia Pacific, and South America.
Teams currently evaluating whether to go direct to Anthropic or OpenAI versus routing through a cloud provider should factor the total organizational switching cost: not just API compatibility, but project management, usage analytics, and the agent-routing infrastructure the new Bedrock console bundles together.
AWS blog (aws.amazon.com/blogs/aws), published June 5, 2026.
