Exa just shipped Websets V2, and the pitch is simple: build a big list from the web faster, find more of what you asked for, and pay less. The company says datasets come together 4x faster, matches are up 20%, and the new usage pricing works out to roughly half the cost per result. The announcement went up on 17 November.
What Websets does
If you haven’t used it, Websets is Exa’s list builder. Type what you’re after, say every protein-design paper that uses diffusion models and releases its code, and it goes hunting through live web data plus Exa’s own index. That index covers more than 1 billion people, 70 million companies, 1.2 billion products and 500 million research papers. Every row comes back with its sources and a note on why it qualifies. Want more detail? Add a column for a repo link, a lab’s focus or a company’s latest project, then export the lot.
What changed in V2
Speed, mostly, and that 4x figure is Exa’s. Accuracy and recall are both up 20%, and Websets now finds 20% more matches on average. The subscription is gone too. Usage comes out of your existing Exa credits, and anyone on the usage plan gets the API.

What it costs
Pricing is where I’d spend my attention. You’re billed for two things. The first is compute, in Agent Compute Units, which is the model doing the thinking: reading your query, judging candidates, checking them against your criteria and filling columns. The second is searches. One query can fire off a lot of them. It all draws from the same credit balance as the Search API.
| Item | Price |
|---|---|
| Agent Compute Unit | $0.10 |
| Searches | $5 per 1,000 ($0.005 each) |
| Contact enrichment, per email found | $0.02 on top |
| Contact enrichment, per phone number found | $0.07 on top |
Here’s Exa’s own example. A run that burns 3 compute units and 250 searches costs $1.55. Get 50 verified matches out of it and each one costs about $0.031. Exa is upfront that there’s no fixed price per match. Tougher queries and heavier enrichment simply use more compute.

Switching plans
New accounts land on usage pricing straight away. If you’re on Starter or Pro, V2’s speed and quality gains show up on their own, but your old pricing and separate Websets credits stay put until you switch. Switching ends the legacy subscription for good. There’s no undo.
- Open the Websets billing page and pick your team.
- Choose Upgrade to new pricing.
- Read the preview of how your remaining eligible Websets credits convert to Exa credits.
- Confirm.
How it stacks up
Exa ran two benchmarks, Company FindAll and People FindAll, with 20 queries each. A result only counts if it genuinely meets the criteria. On companies, Websets returned 514 verified matches, or 25.7 a query, at $0.024 each. On people it returned 613, or 30.65 a query, at $0.0169 each.
| Eval | Measure | Websets | Parallel FindAll | Claude Code, Opus 4.8 | Gemini 3.1 Pro |
|---|---|---|---|---|---|
| Company | Verified matches per query | 25.7 | about 25.5 | about 18 | about 13.4 |
| Company | Price per match | $0.024 | about $0.28 | about $0.16 | about $0.02 |
| People | Verified matches per query | 30.65 | about 27 | about 10.4 | about 6.2 |
| People | Price per match | $0.0169 | about $0.28 | about $0.25 | about $0.06 |
Two things jumped out at me. On the company list, Parallel’s FindAll lands about the same number of matches but costs roughly 11 times more per match. Gemini 3.1 Pro is the only rival priced anywhere near Websets, and it verifies about half as many. On the people list, Websets wins on both counts. One caveat is plain fact: Exa ran these evals itself, and the charts show aggregate results with no confidence intervals.

Where I’d start
For anyone building lists at volume, the new pricing matters more than the speed-up. Cost now follows compute and searches rather than a plan tier. Before you switch, work out your own number: compute units times $0.10, plus searches times $0.005, divided by verified matches. Read the credit conversion preview closely, because you can’t reverse it. And only add enrichment columns you’ll use, since each one spends compute and searches too.
Sources
Discover more from Sulat AI & Tech News
Subscribe to get the latest posts sent to your email.




