Responses API Bundles Web Search

Bundled search turns web retrieval from a separate vendor decision into a default feature of the model layer. Once OpenAI can call live web search inside the same Responses API request, a developer no longer has to stitch together a model, a search API, result parsing, and citation handling. That pushes independent search providers out of the default path and forces them to win on cases the model labs still do poorly, like deeper research workflows, higher recall, or domain specific data access.

• This is not just a product feature, it changes who owns the developer relationship. OpenAI launched Responses API with built in tools including web search in March 2025, and frames it as a simpler way to build agents without integrating multiple APIs or vendors. That is classic value chain compression.
  1 openai 2 openai
• The same pattern is visible across the stack. Google lets Gemini ground on Google Search and on Vertex AI Search data stores, while Microsoft retired Bing Search APIs on August 11, 2025 and directs customers to Grounding with Bing Search inside Azure AI agents. Search is moving from standalone API to bundled agent capability.
  3 google 4 google 5 microsoft
• Independent providers still matter where generic bundled search is not enough. Internal interviews show Exa winning on large result sets and full text extraction for data pipelines, Parallel winning on longer agentic research runs, and buyers often viewing service, latency, and workflow fit as the main differences because baseline search quality is converging.
  6 sacra 7 sacra 8 sacra 9 sacra

The next phase is a split market. Generic web search will increasingly be free or near free inside foundation model platforms, while surviving independents move up the stack into research agents, proprietary connectors, and domain specific knowledge bases for areas like finance, law, and medicine where bundled web search is still too shallow.