Benchmark explainer
What is DeepResearch Bench?
Multi-step deep-research agent capability on PhD-level research tasks across 22 academic domains.
DeepResearch Bench as reported in Epoch AI's Capabilities Index CSV.
How it's scored
- Metric
- normalized performance in Epoch CSV, rendered as accuracy (%)
- Score ceiling
- 100
- Construction
- 100 expert-authored PhD-level tasks split 50/50 English/Chinese; domain distribution guided by 96k+ real user-query analysis from a search-enabled LLM.
- Human baseline
- Calibrated against expert human report quality on RACE dimensions (paper §5.3).
How trustworthy
Discrimination
50.5/100Does it still separate models?
Saturation headroom
90.8/100How far from ceiling / clustered at the top?
Contamination resistance
25/100Public vs held-out; training-leak risk.
Harness comparability
100/100Is it apples-to-apples, or mixed / vendor-optimized?
Freshness
93/100How old is the benchmark?
Contamination history: Released 2025-06 with public task prompts; web-search-enabled evaluation exposes models to potentially pre-seen material.
Limitations: LLM-as-judge evaluation (RACE) inherits LLM biases; FACT citation metric assumes cited sources are verifiable; task set Chinese/English only.
Who leads DeepResearch Bench
| Model | Score | Evidence |
|---|---|---|
| Claude Opus 4.6 | 55.3 | unverified |
| Claude Sonnet 4.6 | 54.9 | unverified |
| Claude Opus 4.5 | 54.8 | unverified |
| GPT-5.5 | 54 | unverified |
| Claude Sonnet 4.5 | 52.6 | unverified |
Compare the top DeepResearch Bench scorers
Frequently asked questions
DeepResearch Bench: DeepResearch Bench as reported in Epoch AI's Capabilities Index CSV.
Claude Opus 4.6 leads DeepResearch Bench at 55.3. The full leaderboard above lists every recorded measurement, not just the headline number.
24 models have recorded scores on DeepResearch Bench, spanning a score spread of 20.2.
tensor.news grades DeepResearch Bench B for integrity (score 70/100), ranking #54 of 65 benchmarks we assess across discrimination, saturation, contamination resistance, harness comparability, and freshness.
No — DeepResearch Bench still has headroom and continues to discriminate between models rather than bunching them at the ceiling.
Its test set is unknown, so contamination risk is on the table: a high score may partly reflect training-data overlap rather than capability.
Scores are reported under a consistent harness, so comparisons on DeepResearch Bench are reasonably apples-to-apples.
Every DeepResearch Bench measurement is source-backed and tagged reproduced, self-reported, or unverified; the underlying record cites arxiv.org/abs/2506.11763.