Exa 3.0 launched: independent evaluation says it ties Perplexity,Token usage reduced 47%

Article

Exa 3.0 Search Token Reduced consumption 47%, ties Perplexity in independent evaluations

Dongcha Beating AI Flash,AI Search company Exa has released Exa 3.0, primarily providing AI Agent with web-connected search capabilities. Exa claims that the new version ranked first in all 8 evaluations covering deep research, code retrieval, financial and corporate research, surpassing Perplexity, Brave, and Parallel in the tests.

Third-party organization Artificial Analysis tested using the same model and Agent framework and found that Exa Auto achieved an overall score of 80, tying for first place with Perplexity Search. However, Exa took an average of 43.7 seconds to complete an individual task, while Perplexity required only 25.5 seconds. The lower-priced Exa Lite scored 79, with completion time reduced to 29 seconds.

The focus of this upgrade is reducing the cost of AI reading search results. The new Dynamic Highlights no longer extracts content page by page; instead, it analyzes all search results collectively, allocating more Token to the most valuable excerpts while removing duplicate information. Exa says that, compared with reading full webpages, this can reduce 95%consumption of Token. Artificial Analysis testing found that, compared with the previous summary mode, the Agent's per-task Token total fell from 299000 to 157000, a reduction of 47%, while total cost declined by 21%. This technology had been available in preview as early as August and is now officially launched.

Exa has also trained a larger Embedding model and expanded its search index, which now covers 1270000000 e-commerce product listings, more than 211000000 merchants, and real-time data such as flights, weather, and sports events. The newly launched Lite tier charges 4 USD per searches, while Auto costs 7 USD; both are now available through API.

Original link https://m.theblockbeats.info/flash/372742

Key points

AI Search company Exa has released Exa 3.0, primarily providing AI Agent with web-connected search capabilities.
Exa claims that the new version ranked first in all 8 evaluations covering deep research, code retrieval, financial and corporate research.
The report says that Artificial Analysis tested using the same model and Agent framework, and Exa Auto's overall score was 80, tying Perplexity Search for first place.
The report says that Exa took an average of 43.7 seconds to complete an individual task, compared with 25.5 seconds for Perplexity; Exa Lite scored 79 and took 29 seconds.
Artificial Analysis test data says that, compared with the previous summary mode, the Agent's per-task Token total fell from 299000 to 157000, a reduction of 47%, while total cost declined by 21%。
Exa says that Dynamic Highlights compared with reading full webpages, it can reduce 95%consumption of Token.
Exa Lite charges 4 USD per searches, while Auto costs 7 USD; both are available through API.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

Exa 3.0 seeks to enable AI Agent to read less duplicate or lower-value content after web-connected searches and concentrate Token on more valuable excerpts. The test results in the report show that Exa Auto's overall score matched Perplexity Search's, but it took longer to complete tasks; the Lite tier scored slightly lower but took less time.

Why it matters to readers

For AI Agent using web-connected search,Token usage, cost, score, and response time are all metrics to consider when selecting a search service. This report shows that Exa has upgraded its approach to reducing Token and offering different pricing tiers.

Beginners can understand Token as AI a unit of measurement used when processing text.Token Using less may reduce costs, but search quality and task-completion time should also be considered.

Risks and unknowns

The evidence does not provide a link to Artificial Analysis's original evaluation report, so the testing methodology and complete results cannot be verified.
8 First place in all 95% Token evaluations, the reduction magnitude, index coverage scale, and other data are presented in the report as Exa's claims.
The evidence does not specify whether the evaluation covers long-term stability, different query types, or actual production environments.
The evidence does not provide responses from Perplexity, Brave, or Parallel regarding the evaluation results.

Exa 3.0 launched: independent evaluation says it ties Perplexity,Token usage reduced 47%

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center