Anthropic adds web search to its API, starting at $10 per 1,000 searches, giving Claude 3.7 Sonnet, 3.5 Sonnet, and 3.5 Haiku access to up-to-date information
Anthropic is launching a new API that allows its Claude AI models to search across the web. Developers using it can build Claude-powered apps …
Context & Ripple Effects
Anthropic first made web search available in a preview for paid Claude users, and had already added document-grounding citations to its API. This release extends the fresh-information capability from the Claude product into developers’ applications.
The feature lands across Sonnet and Haiku variants, spanning higher-capability and latency-sensitive use cases rather than treating web access as a standalone consumer feature.
First-order effects
- Developers can now invoke web search alongside Claude 3.7 Sonnet, 3.5 Sonnet, and 3.5 Haiku through Anthropic’s API, allowing applications to use current web information rather than only model knowledge or supplied documents.
- Search becomes a separately metered input at $10 per 1,000 searches, adding a visible per-query cost to Claude application design.
Second-order effects
- Teams building Claude products will need to decide which requests justify a paid search call, balancing freshness against the existing model and token costs—especially for high-volume workflows associated with Haiku’s latency-sensitive positioning.
- The API makes Anthropic a more complete application-layer provider: customers can combine model inference, retrieved web material, and source-grounding features within one vendor workflow.
Third-order effects
- If model vendors continue bundling retrieval and grounding into their APIs, competition will shift from model quality alone toward the cost and reliability of an end-to-end answer stack.
- As AI applications draw more directly on web content, publisher access and attribution rules may become a more material constraint on answer-engine products; the corpus does not establish how Anthropic will handle those controls.
The trend: Foundation-model APIs are evolving into integrated answer-generation platforms that package inference, retrieval, and grounding as metered services.