Exa Launches Agent Ultra: A Subagent Swarm Deep Research API Built for Exhaustive List Building
Exa launched Agent Ultra, a hosted subagent research API it says leads four deep-research benchmarks.
Exa released Agent Ultra, the highest-effort mode of its hosted Exa Agent API, for exhaustive list building, entity enrichment, and multi-source research. It splits tasks across parallel subagents and mixes frontier and faster models, invoked by setting effort to ultra. Exa’s vendor-reported results are WANDR soft recall 81.4%, DeepSearchQA F1 93.9%, WideSearch row-level F1 58.9%, and 2,451 passing entities per Company Find-All task, ahead of Opus 5.5, GPT-6 Astra, and Perplexity Agent. Typical runs take about 30 minutes, very hard tasks up to three hours, at up to $20 per run by default.
- Agent Ultra is a hosted API mode and is not open weights.
- Exa claims leads on WANDR, DeepSearchQA, WideSearch, and Company Find-All.
- Company Find-All averaged 2,451 entities versus 146 for Opus 5.5.
- Runs default to a $20 cap, adjustable from $1 to $100.
- Scores are vendor-reported and not independently reproduced.
- Vendors
- Exa
- Products
- Exa Agent APIAgent Ultra
- Organizations
- ExaAnthropicOpenAIPerplexityGoogle DeepMind
- AI models
- Opus 5.5GPT-6 Astragpt-6-luna
From the original article
4 min read at marktechpost.com
Exa has released Agent Ultra, the highest effort mode of its Exa Agent API. It coordinates subagents across thousands of sources for list building and entity enrichment. Exa reports it beats Opus 5.
By Michal Sutter · marktechpost.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.