The Most Brutal AI Benchmark Ever Made #benchmark #ai #perplexity

The Most Brutal AI Benchmark Ever Made #benchmark #ai #perplexity

Perplexity's Open Source Benchmark: A Challenge for Research Agents

Overview of the Benchmark

  • Perplexity has released a benchmark that exposes the limitations of research agents, with its best score still being inadequate.
  • The benchmark consists of 500 real-world tasks derived from actual research jobs, including due diligence and market analysis.

Task Complexity

  • Each task requires agents to find information on 50 companies, detailing every employee and providing evidence for each fact.
  • The median task demands 50 entries backed by 245 records, totaling approximately 170,000 records across all tasks.

Evaluation Criteria

  • The benchmark does not have a fixed answer key; it evaluates claims based on the accuracy of cited pages.
  • Agents struggle primarily in two areas: discovery (finding all required information) and evidence (supporting claims with accurate citations).

Performance Insights

  • Perplexity's search agent leads with a soft score of 0.36, while Anthropic follows at 0.25; other systems score below 0.13.
  • Even leading systems incur costs exceeding $5 per task, indicating widespread challenges in achieving satisfactory performance.

Scoring Challenges

  • In terms of hard scores requiring perfect accuracy, even top performers succeed only about one in seven times.
  • Perplexity notes that while partial progress is common, complete coverage remains elusive.

Future Implications

  • The grading pipeline also serves as a training data source for improving agent performance based on identified weaknesses.
  • The open-source nature of this benchmark raises questions about its integrity—whether it's an honest assessment or merely a showcase by its creators.
Video description

WANDR is Perplexity's new open-source AI research agent benchmark — and every LLM agent, including Perplexity's own, scores terribly on it. 500 real research tasks, 170,495 records to find and prove, and the best soft-F1 score is just 0.36. ---- 🚀 DYNAMOUS AI COMMUNITY Want to learn agentic coding with live daily events and workshops? Check out Dynamous AI: https://dynamous.ai/?code=646a60 Get 10% off here 👉 https://shorturl.smartcode.diy/dynamous_ai_10_percent_discount ⚡ HOSTINGER — RELIABLE HOSTING FOR YOUR PROJECTS (10% OFF) Whether you're shipping a portfolio, a side project, n8n flows, or AI agents — I use Hostinger for fast, affordable VPS + web hosting. Get 10% off here 👉 https://hostinger.com/DIYSMARTCODE (Affiliate link — costs you nothing, supports the channel.) ---- What you will see in this 3-minute breakdown: - WANDR: the open-source benchmark that breaks deep + wide research agents - Why 500 real tasks (due diligence, literature review, market analysis, talent sourcing) = 170,495 records to dig up and prove - The grading trick: no fixed answer key — the grader re-opens every cited page and checks each claim against the live evidence - The two walls every LLM agent hits: discovery (can't find them all) and evidence (up to 68% of cited pages don't support the claim) - The leaderboard: Perplexity 0.36, Anthropic ~0.25, everyone else under 0.13 — and $5+ per task - The hard score: even the leader earns full credit on just 1 in 7 of the records a task needs - The twist: the same grading pipeline doubles as a training-data factory WANDR benchmark (Perplexity Research): https://research.perplexity.ai/articles/wandr-benchmark-evaluating-research-agents-that-must-search-wide-and-deep Perplexity on X: https://x.com/perplexity_ai A benchmark nobody can beat yet — is that the most honest benchmark out there, or just a flex from the lab that built it? Drop your take below. #WANDR #Perplexity #AIAgents #LLM #AgenticAI #AIResearch #ResearchAgents #DeepResearch #LLMBenchmarks #AIBenchmark #MachineLearning #LLMAgents #AutonomousAgents #AIEvaluation #Anthropic #OpenSource #AISearch #InformationRetrieval #AICoding #DeepLearning #ArtificialIntelligence