Introducing Research 2.0—our most comprehensive research agent yet.

Our latest agent Research 2.0, has been published to all Surf users.

It is a full rebuild of the system behind Surf's Research agent. Response depth has increased by 2x, hallucination rate dropped 4x, and the agent now has access to 37 new API endpoints across prediction markets, capital flows, and DeFi data that didn't exist in v1.5. This upgrade significantly enhances the user research experience in two main dimensions: completeness and reliability.

We spent Q1 building Surf's foundational products — Studio, Skills, Pulse. With that infrastructure in place, we turned back to Research: the product that serves 500K+ users and is still growing. We rebuilt the orchestration layer from scratch — what data the agent pulls, how many sources it cross-references, how it structures the answer. The goal is to set a new standard for AI-driven crypto research. This is where we are so far.

We evaluated thousands of real user queries — sampled from actual Surf traffic, not synthetic prompts. Each output was graded by an LLM judge, checking whether claims are backed by data the agent actually retrieved through tool calls. v1.5 and v2.0 were tested on comparable random samples from the same query pool.

The answers are more reliable. Hallucination rate dropped from 32% to 8% — a 4x reduction. Faithfulness went from 72% to 85%. Completion hit 97%. When Research 2.0 gives you a number or a conclusion, it's traced back to real data the agent actually retrieved. You can act on it without opening five tabs to verify.

The answers are more complete. Report logic consistency jumped from 55% to 94% — the single biggest improvement. In v1.5, more than 4 in 10 outputs contradicted themselves at some point. v2.0 holds together end to end. Relevant evidence went from 60% to 82%, meaning the agent pulls the right data for the question you actually asked, not just adjacent data it happened to find. The result is an answer you can use as-is, not one you have to mentally reassemble.

Research 2.0 added 37 new data connections to the agent's toolkit across three categories that v1.5 was completely blind to.

Polymarket & Kalshi. Research can now pull live and historical data from Polymarket and Kalshi — odds, probability movements, settlement outcomes, and cross-platform pricing gaps. v1.5 couldn't answer a single prediction market question.

Institutional capital flows. Fund holdings, ETF flows, token launches, and exchange listing events. You can now ask Research where institutional money is moving and how it correlates with price action.

DeFi comparisons. v1.5 could look up one protocol at a time. v2.0 can compare yield across protocols, rank projects by onchain activity, and track bridge volume across chains.

Hyperliquid. Dedicated coverage including funding rates (full history back to genesis), open interest, whale positions, and price data down to 1-minute candles.

Uses the new Polymarket endpoints to analyze settlement patterns at scale — scanning 30 days of 5-minute markets for probability reversals that signal mispricing. The kind of analysis that previously required building a custom scraper and doing manual data work.

Full report here

Cross-references listing data, market cap timelines, token creation dates, and ecosystem relationships across multiple data sources in one query. Pulls from both market data and project intelligence endpoints to build a full picture for each token.

Full report here

"Analyze the following three tokens and identify whether there are any common wallets that bought more than $1,000 worth of each token at an early stage and later made a profit.

Cross-token wallet intersection. Given three separate tokens, Research identifies addresses that bought early and profited on all three, the pattern that separates informed wallets from lucky ones. The kind of multi-contract analysis that takes hours of Dune queries and spreadsheet work, answered in a single prompt.

Full report here

Research 2.0 is a major step, but it's a fraction of the full picture of Surf 2.0. The team is continuing to build and iterate across the entire product line.

The next milestone: a personalized research experience. We're building the option to connect your portfolio so the agent understands your positions and trading style. Instead of generic answers, Research will tailor its analysis to what you actually hold and how you trade.

The goal hasn't changed — make Surf the default AI chatbot for all crypto research. Research 2.0 gets the foundation right. What comes next makes it yours.

Stop looking for answers. Just ask surf.