Snowflake AI Research used Tavily Search API in ArcticSwarm’s live-web BrowseComp evaluation. The search router tried Brave first, followed by Tavily and then Serper. The paper reports that Tavily answered 5,610 of 17,125 unique outbound search queries, or 32.8%. A shared cache deduplicated searches across subagents within each case, achieving a 92.4% search-cache hit rate across 226,213 search tool calls. This disclosure concerns a research evaluation.
Separately, Tavily announced a Snowflake Intelligence integration and a native app on Snowflake Marketplace. These let Snowflake users add web search to agents and worksheets. Tavily’s documentation describes installing the app, configuring external API access and attaching its search procedure to an agent.
Snowflake AI Research’s ArcticSwarm evaluation routed live-web searches through Brave, then Tavily, then Serper. Tavily answered 5,610 unique outbound queries, or 32.8%. A shared per-case cache deduplicated searches across subagents and achieved a 92.4% search-cache hit rate across 226,213 search tool calls.
ArcticSwarm research paper15 Oct 2026 · arXiv · Web pageSnowflake AI Research’s ArcticSwarm evaluation routed live-web searches through Brave, then Tavily, then Serper. Tavily answered 5,610 unique outbound queries, or 32.8%. A shared per-case cache deduplicated searches across subagents and achieved a 92.4% search-cache hit rate across 226,213 search tool calls.
Tavily Snowflake native-app documentation
15 Oct 2026 · Tavily · Web page
Tavily documents installing its Snowflake native app and configuring web search for worksheets and Snowflake Intelligence agents.
Tavily Snowflake native-app documentation15 Oct 2026 · Tavily · Web pageTavily documents installing its Snowflake native app and configuring web search for worksheets and Snowflake Intelligence agents.
Snowflake provides a managed cloud platform for enterprise data, analytics and AI. Data engineers, analysts, data scientists and application developers use its AI Data Cloud to build data pipelines, query and share information, develop applications and run machine-learning workloads. Its products include data warehousing and analytics, Cortex AI for language-model and search applications, Snowpark for processing data in Python, Java and Scala, and Horizon for centralized governance. Customers span financial services, healthcare, retail, manufacturing, telecommunications and the public sector.
Snowflake makes Iceberg v3 table upgrades generally available
Customers can upgrade existing Iceberg v2 tables to v3 in place while preserving names, privileges and downstream references. Snowflake also made expression defaults generally available for its managed Iceberg v3 tables.
Cloud platform for enterprise data, analytics and AI. · Infrastructure
How they use Tavily
Snowflake AI Research used Tavily Search API in ArcticSwarm’s live-web BrowseComp evaluation. The search router tried Brave first, followed by Tavily and then Serper. The paper reports that Tavily answered 5,610 of 17,125 unique outbound search queries, or 32.8%. A shared cache deduplicated searches across subagents within each case, achieving a 92.4% search-cache hit rate across 226,213 search tool calls. This disclosure concerns a research evaluation.
Separately, Tavily announced a Snowflake Intelligence integration and a native app on Snowflake Marketplace. These let Snowflake users add web search to agents and worksheets. Tavily’s documentation describes installing the app, configuring external API access and attaching its search procedure to an agent.
Snowflake AI Research’s ArcticSwarm evaluation routed live-web searches through Brave, then Tavily, then Serper. Tavily answered 5,610 unique outbound queries, or 32.8%. A shared per-case cache deduplicated searches across subagents and achieved a 92.4% search-cache hit rate across 226,213 search tool calls.
ArcticSwarm research paper15 Oct 2026 · arXiv · Web pageSnowflake AI Research’s ArcticSwarm evaluation routed live-web searches through Brave, then Tavily, then Serper. Tavily answered 5,610 unique outbound queries, or 32.8%. A shared per-case cache deduplicated searches across subagents and achieved a 92.4% search-cache hit rate across 226,213 search tool calls.
Tavily Snowflake native-app documentation
15 Oct 2026 · Tavily · Web page
Tavily documents installing its Snowflake native app and configuring web search for worksheets and Snowflake Intelligence agents.
Tavily Snowflake native-app documentation15 Oct 2026 · Tavily · Web pageTavily documents installing its Snowflake native app and configuring web search for worksheets and Snowflake Intelligence agents.
Snowflake provides a managed cloud platform for enterprise data, analytics and AI. Data engineers, analysts, data scientists and application developers use its AI Data Cloud to build data pipelines, query and share information, develop applications and run machine-learning workloads. Its products include data warehousing and analytics, Cortex AI for language-model and search applications, Snowpark for processing data in Python, Java and Scala, and Horizon for centralized governance. Customers span financial services, healthcare, retail, manufacturing, telecommunications and the public sector.
Snowflake makes Iceberg v3 table upgrades generally available
Customers can upgrade existing Iceberg v2 tables to v3 in place while preserving names, privileges and downstream references. Snowflake also made expression defaults generally available for its managed Iceberg v3 tables.
