What Tavily shipped: August 2026. A #1 search benchmark, Rox cuts account research from a week to seconds, get the Financial Services & Insurance Kit, Tavily's inside the Nebius Builder Program, where Tavily has been and where its team is headed next. Tavily Team Aug 31, 2026 Tavily now ranks #1 on SealQA and SimpleQA. Language models keep getting better at reasoning, which means, the limiting factor has moved from the model to the quality of the information it receives. So, the Tavily team spent the past several months overhauling core parts of its search system, including reranking and evidence-quality modeling; evidence deduplication and contradiction handling; and index coverage and temporal freshness. Now, Tavily ranks #1 on SealQA-Hard, SealQA-0, and SimpleQA Verified. Tavily wrote out why Tavily made these improvements, the evaluation method Tavily used, plus when and how they help your agents. How Rox cut account research from a week to seconds. Rox helps the Global 2,000 automate their entire revenue cycles. But, to make that happen, their agents need to work off of what's happening right now, and need to be trusted to work across thousands of accounts, fast. Multiple providers were tested, but ultimately, Rox chose Tavily based on cost, a good working relationship with the Tavily team, and unmatched results in their own evals. Tavily now handles millions of Rox's search and extraction requests a day, each answered in milliseconds, so a fresh signal lands on a rep's screen in a few seconds. Tavily joins the Nebius Builder Program. The Nebius Builder Program is now live in early preview. Builders who join the program can learn the Nebius, Tavily, and Token Factory stack, and gain access to the Builder community. When you join, you get Tavily and Token Factory credits, office hours with Nebius engineers, and can take the Nebius Academy certification. If you're looking to try a model, explore inference, add web grounding, take part in training, or learn how the Nebius stack works, this free-to-join program is a must. A starter kit for financial AI agents. Choosing the right retrieval layer for agents in investment research, AML, and risk workflows comes with its own challenges in compliance and audits. On top of delivering accurate findings, teams need to know their web search API is built for secure, auditable retrieval. So, its team put together a run through on how to choose the right provider, and built a starter kit for teams in the finance industry. The Financial Services & Insurance Kit includes a quick-start guide and reference architecture to help you get started with building agents for compliance intelligence, investment research, and merchant risk. New integrations: Convex, OpenCode, and NanoClaw. Convex. Add real-time search and content extraction to a Convex app as an isolated Convex component with tavily/convex-tavily. Your app calls a typed TavilyClient; the component keeps your API key in typed Convex environment config, and owns no database tables of its own. Learn more OpenCode. Give your agents reliable web search and extract with the opencode-tavily plugin in OpenCode, the open-source terminal coding agent. Once installed, your agent can run commands directly from the terminal, with all output routed to a .tavily/ directory so it doesn't flood context. Learn more NanoClaw. Your lightweight, persistent NanoClaw agents in Slack, Telegram, and the CLI can now Search and Extract the live web with no API key required using the new tavily integration. You can run /add-tavily-tool in NanoClaw providers that support MCP, like Claude, Codex, and OpenCode, and install it on multiple agent groups. Learn more On the road. The Tavily team was global in August, supporting hackathons; attending, hosting, and co-hosting events. Evan Rimer from its FDE team gave a talk hosted by Stripe Developers on why web search built for humans doesn't hold up for agents: humans optimize for clicks, agents need to optimize for the next decision. Watch the talk Its team also showed up at an IBM event this month, and caught up with builders around Fenway Park and Santa Clara. Where to find Tavily next. * LangChain Interrupt: NYC - September 24, at Halo in NYC's Financial District. Get tickets | * HumanX Amsterdam - September 22-24, at the RAI. Learn more | * The AI Conference - September 29-October 1, at Pier 48, San Francisco. Learn more | * The Summer Lock-In: London - Tavily is a technology partner, alongside OpenAI, Fal, and Pioneer for this hackathon. RSVP | * Builders & Brews & more hackathons - its recurring builder nights keep rotating through SF, Berlin, and London. Drop in, build something, and meet the team, or join a hackathon. Check events -> Loving tavily? Tell the world. Real reviews from real users carry more weight than anything Tavily could say ourselves. If Tavily has made your work easier, two minutes of your time could be exactly what someone else needs to find the right tool. Connect with Tavily. Questions or feedback on any of this? Tavily is on Discord, X, and LinkedIn.

Tavily Ranks #1 on SealQA and SimpleQA. Tavily is now #1 in accuracy on SealQA-Hard, SealQA-0, and SimpleQA after rebuilding core parts of Search, including evidence ranking, contradiction handling, and index coverage. Here's how and why Tavily did it. Aug 11, 2026 Over the past several months, Tavily has overhauled core parts of its search system: reranking and evidence-quality modeling, evidence deduplication and contradiction handling, and index coverage and temporal freshness. The results: Tavily now ranks #1 on both SealQA-Hard and SealQA-0, outperforming Exa Auto, Parallel Advanced, Perplexity Search, Brave, and You.com. Tavily also topped SimpleQA Verified, though if you're building agents, SealQA is the benchmark worth paying attention to. More on that below. Why SealQA? SimpleQA has overly straightforward, factual questions that are far from real-world agentic search (ranked #1 there too). BrowseComp is at the other extreme. It is so agentic that it hides the actual search quality behind the harness. SealQA strikes the balance: it targets more complex queries while still revealing the underlying search quality in a single step. What was improved? These gains reflect several months of targeted engineering work across three components of the retrieval pipeline: * Reranking and evidence-quality modeling: Tavily substantially improved the post-retrieval scoring functions that rank both sources and the evidence extracted from them. Tavily has long optimized not only for retrieving authoritative, relevant pages, but for identifying the evidence most likely to answer a query. The latest improvements refine how source authority, credibility, query relevance, and evidence quality are weighted together. This is especially important for agentic search: traditional web search helps humans find the right page, while machine-oriented search must also surface the right evidence from that page. * Evidence deduplication and contradiction handling: Tavily improved how the system identifies semantically redundant, overlapping, and potentially conflicting snippets. Reducing repeated or inconsistent evidence increases the information density in the retrieved context and lowers the likelihood that downstream models hedge, overgeneralize, or produce unsupported conclusions. * Index coverage and temporal freshness: Tavily expanded index coverage and improved update responsiveness for rapidly evolving topics. This reduce evidence gaps and the retrieval of stale information, particularly for time-sensitive queries where source recency is a critical relevance signal. Evaluation methodology. Tavily wanted to measure the pure search quality, not hidden behind a complex harness or improved through multiple agentic steps. So, Tavily kept the pipeline short and simple: * Query each provider's Search API with max_results=10 * Pass the returned snippets to a reader model (Summarize the answer using gpt-5.4 Mini with reasoning_effort="medium") * Grade the extracted answer using gpt-4.1-mini with the official SealQA grading prompt, adapted from the original SealQA paper Across all providers, reader model, grader model, prompts, and retrieval budget were identical. The only variable is what each provider returns, which is exactly what Tavily wanted to measure. Why this matters. Language models keep getting better at reasoning. Which means, the limiting factor isn't the model, but the quality of the information it receives. Every improvement Tavily has made to Tavily Search starts from the same question: does this help an agent retrieve better evidence and produce a more accurate answer? Tavily is glad to see that question answered so positively in SealQA. That said, Tavily treat any benchmark as one signal. And Tavily know that no public benchmark can represent your needs better than your end-users. If your team is building agents, Tavily encourage you to run evaluations using queries from your own real-world use cases to see how they perform. How to test out the Tavily improvements. Search improvements across its API, SDKs, CLI, and MCP server are already live for all Tavily users. Start building here.