Exa

Come moonlight with us 🌙 Join Gemini + Exa and some of our favorite AI companies for an exclusive AI hack lab in SF. Curated for top AI researchers, engineers and technical founders. The challenge: launch your side hustle in only 12 hours. Bring a research insight, side project or idea you’ve been sitting on. You’ll have 3 focused hours to build, ship and demo - with credits, technical support, challenge tracks and a big prize pool from Gemini, WorkOS, Convex, GMI Cloud, Photon, Mosaic, Vapi, ElevenLabs, AdaL + Apify. The goal: build something technically interesting, commercially promising and worth continuing after the night organized by Build Club. Link to sign up in the comments below!

Exa is now live on the Machine Payments Protocol (MPP), the open standard from Tempo and Stripe Agents shouldn't need accounts or API keys to search the web! With MPP, agents can pay per request with USDC.e on Tempo. We put together a cookbook for building GTM enrichment workflows with MPP, where agents can research and enrich companies while paying for Exa as they go: https://lnkd.in/g5Jxt4hN Docs: https://lnkd.in/gyCs8RMe

Exa is now one of the largest indexes in the world. We serve 80B pages, track 1.4T urls, and are on track to be Google-scale in early 2027. It's difficult to know the size of indexes, but we estimate Google to be ~1T, Bing to be ~500B, and Yandex ~200B. Brave mentioned in April that they're at 40B. Most of the web is trash that can hurt AI outputs, so you actually have to crawl way more than you serve and train models to filter out the trash. In terms of peak QPS, we estimate Bing at roughly 30k/s and Google at ~500k/s. In our case, because agent traffic often requires large fan-out (deep searches can use dozens of sub-searches) and can be spiky (like when AI labs RL with us), we're already starting to provision for Bing-scale traffic. Within 2 years, agents will search at many millions per second. They'll also want comprehensiveness over all data, far bigger than any index today. So the retrieval infra required will be larger than Google-scale in both dimensions.

Exa will be at Ai4 - Artificial Intelligence Conferences in Vegas this week! Here's how you can engage with us: On Wednesday, August 5th - we'll be hosting a speakeasy event with Adobe and Fin (link in comments) On Thursday, August 6 - our CEO Will Bryk will be speaking on a panel titled "The Future of AI Depends on Perfect Search" If you're exploring Exa and would like a demo you can book time to meet us at booth #1364 in the comments below

This is actually true. Every time someone asks me "what's the value prop of Exa?" - I tell them the same thing. It would be hypocritical of me to sell you Exa if we weren't obsessively using it ourselves. And we are. Forget evals. Forget ARR growth curves. Dogfooding is a bigger KPI than any of them. If the team building a product doesn't reach for their own tool every time they need something, nothing on the dashboard is going to save you. At Exa, we live inside Exa. Our own teams and products have now made 2.3 billion search requests through Exa. Last quarter alone: 692 million. Up ~25x from 18 months ago. Every product we've shipped along the way (fast search, deep search, monitors, agent) has driven our own internal usage up. We're the first heavy users of everything we build. Some of what that looks like: <1> Every prospect who books a call triggers an automated research brief in Slack before the rep joins. It runs on our own API - ICP score, recent signals, tech stack, competitive intel, key contacts, etc. <2> Every inbound lead that hits our Salesforce gets auto-enriched by Exa Agent the moment it lands - company data, ICP fit, tech stack, recent signals. By the time a rep opens the record, it's already a proper prospect card. <3> ~20,000 prospects auto-classified into 15 ICP segments and scored by Exa Deep and now every rep starts their day inside it. Exa scoring who Exa should sell to. <4> Our recruiting team is building a Claude skill that uses Exa Agent to source candidates for every open role at Exa. Best candidates, found by Exa itself. We also just launched Exa in Slack (@ Exa) - the same bot we've been dogfooding internally for months to do lookups, follow-ups and list-building without leaving Slack. Two things we shipped externally that came directly from what we use internally: → Exa Skills - the opinionated skills our team packaged from the best practices we use internally. Install: npx skills add exa-labs/agent-skills. Ready workflows for lead gen, recruiting and company research. → Exa GTM Toolkit - a public cookbook of GTM plays on top of Exa. Every prompt in there was tested on our own GTM first. We're the heaviest user of everything we ship.. before customers ever see it. That's what we mean when we say Exa runs on Exa.

We’re excited to launch Web Search Index: a new benchmark comparing how AI agents perform with different web search tools. The core design choice is isolation. We keep the model and setup fixed, and change only the search tool. We also remove domain-specific tools like SEC EDGAR and CourtListener, making general web search the agent’s only outside source. Across 208 expert-authored legal research tasks and 450 financial analysis questions, Claude Fable 5 with Exa topped the overall index at 48.5%. A few takeaways: Exa performed especially well on finance, scoring about 6 percentage points higher than native search after controlling for task difficulty. The biggest finance gaps showed up in retrieval-heavy categories like Market Analysis, Earnings Analysis, and Adjustments. On legal, native search was ahead by less than 1 percentage point, a difference that was not statistically significant. Search tool choice is domain-dependent: stronger retrieval helps most when the task depends on finding the right underlying data or regulatory text. As agents become more capable, the quality of the tools around them matters more. Web Search Index gives a clearer view into how those tools perform end-to-end on real-world tasks.

Today, we're excited to launch state-of-the-art search over a corpus of 300M+ papers using a custom semantic retrieval system designed for this task. As we enter the era of true superintelligence, agents will increasingly benefit from powerful search over the world's scientific knowledge, accelerating scientific discovery. Top biopharma companies, startups like Anara and Elicit, and Harvard Medical School professors such as Omar Abudayyeh and Jonathan Gootenberg are using Exa in their own research and to build products for scientists. We're also releasing two new research paper benchmarks for others to build on top of. In our evals/testing, we found that Exa has far higher accuracy than traditional tools like Google Scholar. https://lnkd.in/guFTWZ8T

Exa 🤝 Pydantic We're excited to see all the new Pydantic AI agents people will build with this new Exa capability!

🤝 Exa x OpenHuman (TinyHumans AI) Exa is now OpenHuman's default search. Your AI searches the web for meaning, not keywords and gets clean results built for agents to reason over. Live for everyone today. https://lnkd.in/geXzp4-4