Anthropic structured output guide

Anthropic's OpenAI compatibility mode is for comparison, not blind production parity.

The compatibility layer is real and useful, but Anthropic documents important differences. If a team expects OpenAI-style strict guarantees through that layer, they can ship false confidence into production.

anthropic openai sdk compatibility strict ignoredclaude openai compatibility structured outputsanthropic strict function calling ignored
Buy Starter Access All comparisons

What breaks

  • Anthropic states that the OpenAI compatibility layer is primarily for testing and comparison.
  • The `strict` parameter for function calling is ignored in that layer.
  • Some unsupported request fields are silently ignored rather than hard-failing.

What Schema Gateway fixes

  • Flag compatibility-mode risk before deploys.
  • Separate 'schema-valid' from 'provider-guaranteed'.
  • Push teams toward native Claude structured outputs when they need hard conformance.

Official references