DeepResearch Bench
Report quality on PhD-level research tasks with web access.
| Category | Agents |
|---|---|
| Metric | Accuracy |
| Source | Epoch AI (evaluator) |
| License | CC BY 4.0 |
| Benchmark page | deepresearch-bench.github.io |
| Status | Scored |
| Weight | 0.7 within Agents |
| Fit | Discrimination 0.23, noise 0.14 |
Results
| # | Model | Provider | Result | Date |
|---|---|---|---|---|
| 1 | Claude Opus 4.6 High | Anthropic | 55.3% | 2026-11-11 |
| 2 | Claude Sonnet 4.6 High | Anthropic | 54.9% | 2026-11-11 |
| 3 | Claude Opus 4.5 High | Anthropic | 54.8% | 2026-11-11 |
| 4 | GPT-5.5 High | OpenAI | 54.0% | 2026-11-11 |
| 5 | Claude Opus 4.5 Low | Anthropic | 53.7% | 2026-11-11 |
| 6 | Claude Opus 4.6 Medium | Anthropic | 53.2% | 2026-11-11 |
| 7 | Claude Sonnet 4.5 Thinking | Anthropic | 52.6% | 2026-11-11 |
| 8 | Claude Opus 4.6 Low | Anthropic | 51.4% | 2026-11-11 |
| 9 | Claude Sonnet 4.6 Low | Anthropic | 50.4% | 2026-11-11 |
| 10 | Claude Opus 4.8 High | Anthropic | 50.2% | 2026-11-11 |
| 11 | Gemini 3 Flash Preview Low | 49.8% | 2026-11-11 | |
| 12 | GPT-5 Low | OpenAI | 49.6% | 2026-11-11 |
| 13 | GPT-5.5 Medium | OpenAI | 49.6% | 2026-11-11 |
| 14 | Claude Opus 4.8 Low | Anthropic | 49.3% | 2026-11-11 |
| 15 | Gemini 3 Flash Preview Minimal | 49.0% | 2026-11-11 | |
| 16 | GPT-5 Minimal | OpenAI | 48.9% | 2026-11-11 |
| 17 | GPT-5.5 Low | OpenAI | 48.7% | 2026-11-11 |
| 18 | GPT-5 Medium | OpenAI | 48.6% | 2026-11-11 |
| 19 | Claude Opus 4.1 | Anthropic | 48.3% | 2026-11-11 |
| 20 | GPT-5 High | OpenAI | 48.1% | 2026-11-11 |
| 21 | Gemini 3 Flash Preview High | 47.9% | 2026-11-11 | |
| 22 | Gemini 3.1 Pro Preview High | 47.8% | 2026-11-11 | |
| 23 | Claude Sonnet 4.5 Low | Anthropic | 47.5% | 2026-11-11 |
| 24 | Claude Opus 4.8 Medium | Anthropic | 47.4% | 2026-11-11 |
| 25 | Grok 4 | xAI | 47.3% | 2026-11-11 |
| 26 | Claude Opus 4 Thinking | Anthropic | 46.8% | 2026-11-11 |
| 27 | Claude Sonnet 4 Thinking | Anthropic | 46.6% | 2026-11-11 |
| 28 | Gemini 3 Pro Preview Low | 46.3% | 2026-11-11 | |
| 29 | Claude Haiku 4.5 Low | Anthropic | 45.5% | 2026-11-11 |
| 30 | o3 Medium | OpenAI | 45.2% | 2026-11-11 |
| 31 | Gemini 3.1 Pro Preview Low | 44.5% | 2026-11-11 | |
| 32 | Claude Sonnet 3.7 Thinking | Anthropic | 43.6% | 2026-11-11 |
| 33 | Gemini 2.5 Pro Preview 06-05 | 42.8% | 2026-11-11 | |
| 34 | GPT-5.1 Low | OpenAI | 42.8% | 2026-11-11 |
| 35 | Gemini 2.5 Pro | 41.5% | 2026-11-11 | |
| 36 | GPT-5.2 Low | OpenAI | 41.1% | 2026-11-11 |
| 37 | Gemini 3.1 Flash Lite Low | 37.3% | 2026-11-11 | |
| 38 | Gemini 3.1 Flash Lite | 36.4% | 2026-11-11 | |
| 39 | GPT-5.4 mini Low | OpenAI | 36.3% | 2026-11-11 |
| 40 | DeepSeek R1 0528 | DeepSeek | 35.1% | 2026-11-11 |
| 41 | GPT-5.4 Low | OpenAI | 35.1% | 2026-11-11 |