Exa launches Agent API at less than half the cost of GPT-5.5 and Opus

• Exa shipped an Agent API that ExaAILabs' launch post describes as a mixed-model web research endpoint for everything from simple enrichments to giant list-building jobs.
• According to ExaAILabs' benchmark claim, the endpoint packages Exa's search engine, token-efficient contents system, and an efficiency-focused harness into a single API.
• Exa's pricing pitch is aggressive: the launch post said "less than half the cost of GPT 5.5 and Opus," while WilliamBryk's launch thread tightened that to roughly 2 to 10 times lower cost for Opus and GPT 5.5 class web research.
• The company tied the launch to named web research benchmarks, with ExaAILabs' benchmark post calling out BrowseComp, DSQA, and WideSearch.
• The API is live now, according to ExaAILabs' availability post, which linked both the start-building page and the blog post.

You can jump straight to the launch post, and Exa spent most of the announcement budget on a specific claim: its harness plus search stack can hit frontier browsing quality without paying frontier model prices. The benchmark list is unusually explicit for a short launch thread, and ExaAILabs' use-case post says the target workloads already include finance and go-to-market research.

The core product move is packaging Exa's own search and content-retrieval stack behind an agent endpoint instead of selling raw retrieval primitives. ExaAILabs' benchmark post says Agent combines an "agent-first" search engine, token-efficient contents, and a harness designed for efficiency.

That framing matters because Exa is not pitching a single new base model. The launch post says /agent orchestrates "a mixture of cost-effective models," which puts the product closer to a research harness with model routing than to a standalone model release.

Exa attached the launch to two kinds of evidence:

• Benchmark names: ExaAILabs' benchmark post cites BrowseComp, DSQA, and WideSearch.
• Cost framing: WilliamBryk's thread says the endpoint reaches Opus and GPT 5.5 quality web research at 2 to 10 times lower cost.
• Task framing: ExaAILabs' launch post leads with web research and list-building, while a follow-up post says the same system also targets financial and go-to-market work.

The interesting part is the combination. Deep research APIs usually show up as premium, slow, expensive surfaces. WilliamBryk's thread instead frames Agent as "close to the cost of a search API," which is a much more direct pitch at teams already paying for search and retrieval infrastructure.

The release was not a waitlist announcement. ExaAILabs' availability post says Agent is available in the API today, with links to start building and the blog post.

That post also clarifies the product boundary: this is an API launch first, not a consumer-facing research app. The announcement language stays focused on builders, benchmarks, and endpoint economics, not on a chat surface or bundled UI.