DeepResearch Bench
PhD-level research briefs written by an agent with web access, graded on coverage, insight and citations.
Long-horizon tasks with tools: terminals, desktops, browsers, whole jobs. One of 8 comparable tests in agentic, a category the Index averages.
Publisher: DeepResearch BenchWhat a model is asked to do
Write a PhD-level research brief with web access, graded on coverage, insight and citations.
For exampleSurvey the last three years of room-temperature superconductivity claims and assess which replications hold up.
Why it matters. A research agent lives or dies on whether its sources are real and its synthesis is right.
The example is original and illustrative, not an item from the dataset.
- Models scored here
- 19
- Who produced the numbers
- Official leaderboard
- Items graded
- n = 100 (1.0% each)
- Best published result
- 55.3%
- License
- CC BY 4.0 (via Epoch AI)
- Direction
- Higher is better
- Scale
- Published as a share of items solved. For the Index, each score is a share of the best published result (55.3%), so the frontier reads 100. Guessing earns nothing on this test, so no chance correction applies.
- Provenance
- Produced by the benchmark's own leaderboard or a third-party evaluator, not by the model's maker.
Snapshot September 8, 2026
Ranking on this test
19 models on DeepResearch Bench
Best published run per model, and how close each one comes to the best published result on this test (the frontier reads 100). Every number links to the publisher that produced it. Internal runs are marked and carry their run details on the model page. One item on this test is 1.0 points, so read gaps smaller than that as noise.
| # | Model | Score | vs frontier | Run | Produced by |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.6Anthropic | 55.3% | 100.0 | high | Leaderboard |
| 2 | Claude Sonnet 4.6Anthropic | 54.9% | 99.3 | high | Leaderboard |
| 3 | Claude Opus 4.5Anthropic | 54.8% | 99.1 | 20251101 · high | Leaderboard |
| 4 | GPT-5.5OpenAI | 54.0% | 97.6 | high | Leaderboard |
| 5 | Claude Sonnet 4.5Anthropic | 52.6% | 95.1 | 20250929 · 2k | Leaderboard |
| 6 | Claude Opus 4.8Anthropic | 50.2% | 90.8 | high | Leaderboard |
| 7 | Gemini 3 FlashGoogle | 49.8% | 90.1 | preview · low | Leaderboard |
| 8 | GPT-5OpenAI | 49.6% | 89.7 | 2025-08-07 · low | Leaderboard |
| 9 | Claude Opus 4.1Anthropic | 48.3% | 87.3 | 20250805 | Leaderboard |
| 10 | Gemini 3.1 ProGoogle | 47.8% | 86.4 | preview · high | Leaderboard |
| 11 | Grok 4xAI | 47.3% | 85.5 | agent: True | Leaderboard |
| 12 | Gemini 3 ProGoogle | 46.3% | 83.7 | preview · low | Leaderboard |
| 13 | Claude Haiku 4.5Anthropic | 45.5% | 82.3 | 20251001 · low | Leaderboard |
| 14 | GPT-5.1OpenAI | 42.8% | 77.4 | 2025-11-13 · low | Leaderboard |
| 15 | Gemini 2.5 ProGoogle | 41.5% | 75.0 | — | Leaderboard |
| 16 | GPT-5.2OpenAI | 41.1% | 74.3 | 2025-12-11 · low | Leaderboard |
| 17 | Gemini 3.1 Flash-LiteGoogle | 37.3% | 67.5 | low | Leaderboard |
| 18 | GPT-5.4 miniOpenAI | 36.3% | 65.6 | 2026-03-17 · low | Leaderboard |
| 19 | GPT-5.4OpenAI | 35.1% | 63.5 | 2026-03-05 · low | Leaderboard |
The best models, ranked here, working inside Theo.
28 of the models on this page run inside Theo today. Theo picks the right one for each step and always shows which engine answered.