Claude Offers OpenAI Compatibility

Anthropic just made it easier for a team using OpenAI’s software tools to try Claude without rebuilding its app first. Anthropic’s new compatibility layer lets developers keep the OpenAI software development kit and change three things instead: the API key, the base web address, and the model name. (platform.claude.com) That sounds small, but for many companies the model is not the hard part. The hard part is the plumbing around it, and Anthropic’s own quick-start example shows the same OpenAI client pointed at ` with a Claude key and a Claude model. (platform.claude.com) Anthropic is also careful about what this is and what it is not. Its documentation says the layer is meant to “test and compare” Claude quickly, and says the native Claude application programming interface is still the better path for production features. (platform.claude.com) The reason is that “compatible” does not mean “identical.” Anthropic says audio input is ignored, prompt caching is not supported through this layer, and system and developer instructions get merged into one opening system message because Claude handles that part differently. (platform.claude.com) It also says strict function-calling rules are ignored in this mode, so tool output is not guaranteed to match a supplied schema the way some developers expect. Anthropic points those teams back to its native interface, where it says structured outputs, citations, extended thinking, and portable document format processing are available. (platform.claude.com) That tells you where the fight is moving. If one company can imitate another company’s basic chat-completions doorway, then the real competition shifts to everything around the doorway: reliability, tool use, safety controls, and the extra services that save engineers time. (platform.claude.com) Anthropic spent the same week pushing exactly those extras. On April 9, 2026, it launched Claude Managed Agents in public beta, a hosted system for running autonomous agents on Anthropic’s own infrastructure instead of making customers build their own sandboxes, state handling, and tool loops. (the-decoder.com) (platform.claude.com) According to The Decoder’s report on Anthropic’s announcement, Notion, Rakuten, and Sentry are already using that managed system for workspace tasks, enterprise agents in Slack and Microsoft Teams, and automated debugging. Anthropic says sessions can run for hours, survive dropped connections, and use built-in tools like bash commands, file operations, web search, and Model Context Protocol servers. (the-decoder.com) So the simple version of this story is not that Anthropic copied a rival’s interface. It is that Anthropic lowered the cost of trying Claude at the bottom of the stack while trying to win the bigger business at the top of the stack, where companies pay for agents, tools, orchestration, and trust. (platform.claude.com) (the-decoder.com)