[GPT-6 Astra Tops Perplexity Research Agent Rankings: 13.5% Higher Than Fable 5.1, 6.1% Cheaper] Breaking AI News from Dongcha: Perplexity tested GPT-6 Astra using its proprietary Agent benchmark, WANDR, and it scored 0.682—the highest among all models tested so far. The average cost per task is $11.98. Compared to Claude Fable 5.1, Astra's score is 13.5% higher, while costs are 6.1% lower; compared to Opus 5, Astra's score is 27% higher, with costs only 3.3% higher. WANDR is specifically designed to evaluate "broad and deep" research tasks. It includes 500 real-world research tasks, requiring Agents not only to find a few answers but to identify all qualifying entities, thoroughly investigate them, verify their identities, and provide each result with a verifiable source. Typical tasks include competitive analysis, due diligence, literature reviews, market analysis, and talent scouting. 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 has significantly pushed the score higher this time, and it wasn’t achieved by simply increasing costs. [Original Link]