Introducing Exa Agent: frontier web research at less than half the cost of GPT 5.5 and Opus. /agent orchestrates a mixture of cost-effective models to complete any web research task, from simple data enrichments to building gigantic lists.

Exa Agent combines our agent-first search engine and token-efficient contents, with a harness designed for efficiency. The result is an API with the best price-performance on the most challenging web research benchmarks: BrowseComp, DSQA, WideSearch, and others.

Agent can be used for a wide variety of important tasks. Below, you can see how it performs on financial and go-to-market use cases.

Exa Agent is available in the API today. Start building: Blog:

a little bit of knowledge makes up for a lot of intelligence

this is chatgpt moment for the websearch world btw

Exa Agent is here and it's unreal.

someone cooked on this

Let’s gooo!!!

🤯🤯🤯

Video designer cooked on this video 🔥

Why is it even half for web search? And you are saying it like a feature. It should be 1/10

exa-cellent search is here!

We love you guys.

Awesome marketing 20k searches per month on free tier is very generous but hard to justify your pricing at scale.. Does the improvement in result quality justify paying 4-5x more than say.. BrightData for most production use cases?

@grok please tell me in great detail what this Exa Agent can do? Make a list after you answer me. The list has to be as comprehensive as possible.

monitoring the situation just got better

Cost claims like this usually come down to how much of the heavy lifting stays on cheaper models versus fallback to frontier. The real test is consistency on hard research tasks.

This is crazy

always exa first!

the web was never built for humans, it was built for agents

it was silly directing this & cloning myself at the same time

Let’s goo!! /agent finds exa-ctly what you’re searching for 😁

I don’t think your mcp has kept up with the api?

Congrats! Would love to explore our synergies sometime as we could both help each other.

all my coding agents default to exa agent when performing web research. I was lucky to beta test this (:

this is actually the first time i’ve seen someone undercut gpt pricing without sacrificing depth, refreshing

Halving the cost of GPT 5.5 is a big claim, but orchestrating cheaper models usually comes with a massive latency tax. Building gigantic lists this way works for background batch jobs, though it still falls short of Opus when you need high-reasoning accuracy on the fly. ⚡

This is huge!!

Would love to try it out for some of our business use cases.

how's the routing deciding which model gets which subtask? that's the part i'm most curious about bc that decision tree is where the cost wins either hold up or fall apart

does it offer iab categorization of domains and also preclassified page types of domains, e.g. login, careers, pricing, news, leadership etc for each domain. so I can tell my agent to run on subset of domains from IAB vertical and on specific page type

成本能压下来，长列表任务就香多了

imo 'mixture of cost-effective models' is doing more work in that sentence than 'frontier'

Frontier level web research at half the cost is the kind of thing that makes you question why you're still paying full price 👀

exa agent's secret sauce is collapsing the value chain for research tasks, not just copying gpt

Benchmarking against GPT 5.5 when it isn’t even public is a wild marketing move. A mixture of cost-effective models works fine for basic data enrichment, but scraping gigantic lists isn't frontier research. Deep synthesis still needs the raw reasoning power of a single heavy model.

this is useful if the trace is as good as the answer. for web research, i care less about another final summary and more about seeing what it searched, trusted, and ruled out.