header-langage
简体中文
繁體中文
English
Tiếng Việt
한국어
日本語
ภาษาไทย
Türkçe
Scan to Download the APP
Partnered With:

GPT-6 Astra Wins Top Spot in Perplexity Research Agent: 13.5% Higher than Fable 5.1 and 6.1% Cheaper

Insightful Beating AI News: Perplexity benchmarked GPT-6 Astra with its in-house Agent WANDR, achieving a score of 0.682, the highest among models tested so far. The average cost per task is $11.98. Compared to Claude Fable 5.1, it scored 13.5% higher with a 6.1% lower cost; compared to Opus 5, it scored 27% higher with only a 3.3% higher cost.


WANDR specializes in testing "broad and deep" research tasks. It consists of 500 real research tasks where the Agent is required not only to find a few answers but to identify all objects that meet the criteria, then research each one, verify their identities, and provide each result with a verifiable source. Typical tasks include competitive research, due diligence, literature retrieval, market analysis, and talent search.


Previously, Fable 5.1 ranked first with a score of 0.601 and a cost of $12.76 per task, while Opus 5 scored 0.537 with a cost of $11.60. Astra significantly raised the score this time, and it was not achieved by simply increasing the cost.

举报 Correction/Report
Correction/Report
Submit
Add Library
Visible to myself only
Public
Save
Choose Library
Add Library
Cancel
Finish