OpenAI Responses API

Latest Audit Result

The docs-only guide correctly demonstrates core declared capabilities (Exa /search with type auto and contents.text via OpenAI Responses API tool-calling, plus direct /responses research compatibility). However, it is outdated: the OpenAI model name gpt-4o should be updated to a current model (gpt-5.4 or gpt-5.5), and the Direct Research note links to now-deprecated /research/v1 docs instead of /search with type deep-reasoning. No deprecated Exa features are actively used in the guide. The /responses endpoint's streaming, structured output, and retrieval-by-ID features are undocumented despite being in the OpenAPI spec.

Missing Features (5)

• DEPRECATED: The guide's Direct Research note (line 522-524) links to the Research API overview at /reference/research/overview, which is explicitly deprecated as of May 1 2026 — should link to /search with type deep-reasoning instead.
• The OpenAI model in tool-calling examples is gpt-4o (lines 73, 123, 255, 297), which is outdated — current OpenAI docs recommend gpt-5.5 as the default frontier model and gpt-5.4 as the cost-efficient choice.
• Document /responses streaming (stream: true) and GET /responses/{id} retrieval — both are in the Exa OpenAPI spec but absent from the Direct Research section.
• Add text.format examples for the /responses endpoint, including json_schema structured output — ResponseCreateRequest supports text.format with schema, name, description, and strict fields.
• Clarify that the Web Search Tool example uses Exa's native POST /search endpoint (via SDK), distinct from Exa's OpenAI-compatible /responses endpoint, to prevent confusion between the two patterns on the same page.