DeepResearch
A long-context deep research agent benchmark involving diverse tool calls. It employs composite scoring criteria to evaluate agents' long-horizon research and synthesis capabilities.
Core Facts
Timeline (last 90 days)
On day 13 of its 30-day search series, Valyu fully demonstrated a real case: how to build a custom DeepResearch workflow for vendor onboarding research and call it through the TypeScript API.
DeepResearch executes multi-step, long-running processes, with both running cost and time far higher than ordinary queries, so codifying the research framework is valuable.
The authors evaluated RLDS on four agent benchmarks: FrozenLake, HotpotQA, ScienceWorld, and DeepResearch.
The improvements of RLDS on HotpotQA and DeepResearch fall within the noise range, consistent with the diagnostic prediction that these two tasks have limited recoverable information.
The DeepResearch task involves four tools including search, summarization, and citation, and uses a composite rubric reward to measure report quality.
All Facts (6)
Deep Research能够自主规划研究路径、在互联网上执行多轮搜索、阅读并综合数十乃至上百个网页内容，最终生成结构化的深度报告
80%UnverifiedDeepResearch executes multi-step, long-running processes, with both running cost and time far higher than ordinary queries, so codifying the research framework is valuable.
50%UnverifiedThe authors evaluated RLDS on four agent benchmarks: FrozenLake, HotpotQA, ScienceWorld, and DeepResearch.
50%UnverifiedThe improvements of RLDS on HotpotQA and DeepResearch fall within the noise range, consistent with the diagnostic prediction that these two tasks have limited recoverable information.
50%UnverifiedThe DeepResearch task involves four tools including search, summarization, and citation, and uses a composite rubric reward to measure report quality.
50%UnverifiedOn day 13 of its 30-day search series, Valyu fully demonstrated a real case: how to build a custom DeepResearch workflow for vendor onboarding research and call it through the TypeScript API.
50%