WANDR

GPT-6 Astra
2026-09-04 07:36:19

GPT-6 Astra Tops Perplexity's Research Agent Benchmark, Beating Fable 5.1 by 13.5% at Lower Cost

Perplexity's WANDR benchmark has ranked GPT-6 Astra as the top-performing research agent model, achieving a score of 0.682 at an average cost of $11.98 per task. This outperforms the previous leader, Claude Fable 5.1, by 13.5% while costing 6.1% less. Compared to Opus 5, Astra scores 27% higher with only 3.3% more cost. The WANDR benchmark evaluates agents on 500 real-world research tasks that require exhaustive identification, verification, and source attribution. Typical tasks include competitive analysis, due diligence, literature review, market analysis, and talent scouting. Previously, Fable 5.1 led with 0.601 at $12.76 per task, and Opus 5 scored 0.537 at $11.60. Astra pushed the score significantly higher without relying on disproportionately higher costs.

30
GPT-6 Astra Tops Perplexity's Research Agent Benchmark, Beating Fable 5.1 by 13.5% at Lower Cost