DRACO
100 deep-research tasks from real Perplexity queries across 10 domains, graded on about 40 expert rubric criteria each.
Built and saturation-tested with Perplexity's own system, which also tops the paper's results. No public code repo found.
In the authors’ words
We present DRACO (Deep Research Accuracy, Completeness, and Objectivity), a benchmark of complex deep research tasks. These tasks, which span 10 domains and draw on information sources from 40 countries, originate from anonymized real-world usage patterns within a large-scale deep research system. Tasks are sampled from a de-identified dataset of Perplexity Deep Research requests, then filtered and augmented to ensure that the tasks are anonymized, open-ended and complex, objectively evaluable, and representative of the broad scope of real-world deep research use cases. Outputs are graded against task-specific rubrics along four dimensions: factual accuracy (accuracy), breadth and depth of analysis (including completeness), presentation quality (including objectivity), and citation quality. DRACO is publicly available at https://hf.co/datasets/perplexity-ai/draco.
- Level
- Expert Specialists with years of training, often over hours of work
- Scored
- Graded by a model A language model scores each response, usually against a rubric.
- Answers
- Open-ended Free-form text, documents, designs or media judged as a whole.
Claude Opus 5 leads with 88.6%
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
- 1 Claude Opus 5 88.6%
- 2 Claude Fable 5.1 87.7%
- 3 Claude Opus 5.5 87.4%
- 4 Claude Sonnet 5.5 87%
- 5
Ling-3.1-flash 85.5% - 6
Step 5 Preview 83.3% - 7 Claude Haiku 5.5 81.5%
- 8 Claude Opus 4.8 80.4%
- 9 Claude Opus 4.7 77.7%
- 10
Tencent Hy4 preview 77.2% - 11
MiniMax-M3 73.2%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5 | 88.6% | Jul 2026 | Model card | |
| 2 | Claude Fable 5.1 | 87.7% | Sep 2026 | Model card | |
| 3 | Claude Opus 5.5 | 87.4% | Sep 2026 | Model card | |
| 4 | Claude Sonnet 5.5 | 87% | Sep 2026 | Model card | |
| 5 | Ling-3.1-flash | 85.5% | Oct 2026 | Model card | |
| 6 | Step 5 Preview | 83.3% | Sep 2026 | Model card | |
| 7 | Claude Haiku 5.5 | 81.5% | Oct 2026 | Model card | |
| 8 | Claude Opus 4.8 | 80.4% | May 2026 | Model card | |
| 9 | Claude Opus 4.7 | 77.7% | Apr 2026 | Model card | |
| 10 | Tencent Hy4 preview | 77.2% | Aug 2026 | Model card | |
| 11 | MiniMax-M3 | 73.2% | Jun 2026 | Model card |
3 models have held number one since its release in Feb 2026
Lab claims count from the day the card came out, independent runs from the model’s release.
The best score passed 85% in Jul 2026
All the scores we have, one dot each, so a model tested at several settings appears more than once.
4 labs put it in a launch post in the last six months
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| 13 Oct 2026 | Ling-3.1-flash | 85.5 | Draco |
| 20 Sep 2026 | Step 5 Preview | 83.3 | Draco |
| 28 Aug 2026 | Tencent Hy4 preview | 77.2 | DRACO |
| 1 Jun 2026 | MiniMax-M3 | 73.2 | DRACO |
7 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| 7 Oct 2026 | Claude Haiku 5.5 system card | 81.5 | DRACO, normalized score at max effort, Claude Opus 4.6 judge, printed on chart (Figure 8.8.2.A) | Capabilities |
| 28 Sep 2026 | Claude Sonnet 5.5 system card | 87 | DRACO normalized score at max effort, 980k token budget (8.11.2, Figure 8.11.2.A) | Capabilities |
| 22 Sep 2026 | Claude Opus 5.5 system card | 87.4 | DRACO normalized score at max effort, 980k-token budget, Claude Opus 4.6 judge, printed on chart (8.11.2, Figure 8.11.2.A) | Capabilities |
| 1 Sep 2026 | Claude Fable 5.1 system card | 87.7 | DRACO normalized score at max effort, 980k-token budget, Claude Opus 4.6 judge, printed on chart (Figure 8.12.2.A) | Capabilities |
| 24 Jul 2026 | Claude Opus 5 system card | 88.6 | DRACO normalized score at max effort, 980k-token budget, Claude Opus 4.6 judge, printed on chart (8.10.4, Figure 8.10.4.A) | Capabilities |
| 28 May 2026 | Claude Opus 4.8 system card | 80.4 | DRACO normalized score, max effort, 1M-token limit with compaction, Claude Opus 4.6 judge (8.10.4) | Capabilities |
| 16 Apr 2026 | Claude Opus 4.7 system card | 77.7 | DRACO normalized score, adaptive thinking at max effort, 1M token limit, Claude Opus 4.6 judge (8.8.4) | Capabilities |
More in Tool use
- AutomationBench In use Agent completes end-to-end business workflows inside a simulated company exposing dozens of REST APIs. Graded on resulting state.
Gemini 4 Carbon 53.9% - MCP Atlas In use Multi-step workflows over real MCP servers. Scores tool-use competency on realistic tasks.
Muse Spark 1.1 88.1% - BFCL V4 In use Models call functions and tools, including multi-turn agentic tasks with web search and memory. Checked against expected calls and outcomes.
Qwen3.7-Max 75% - τ³-bench Banking In use Agents act as banking customer-service reps over many turns with a simulated user, following policy. Pass^1 on the final database state. Qwen 3.8 Max 55.1%
- Claw-Eval In use 300 human-verified agent tasks (service orchestration, multimodal, multi-turn dialogue) audited via traces and snapshots. Rubric scores with Pass^3 consistency. Claude Sonnet 4.6 81.4%
- DeepSearchQA In use Questions that need multi-step search and synthesis. Final answers graded against references. Kimi K3 95%