Public benchmark results for Grep — an AI-powered deep research platform by Parcha.
DRACO — February 2026
Evaluation on Perplexity's DRACO benchmark: 100 open-ended research tasks across 10 domains, scored on 3,934 weighted rubric criteria by Gemini-2.5-Pro.
| Metric | Value |
|---|---|
| Normalized Score | 78.6% |
| Domains Won (vs Perplexity) | 9 / 10 |
| Rubric Axes Led | 4 / 4 |
Grep ranks #1 overall, ahead of Perplexity DR (70.5%), Gemini DR (59.0%), and OpenAI DR o3 (52.1%). Leads all 4 rubric axes: Factual Accuracy (+7.5pp), Breadth & Depth (+7.2pp), Presentation (+3.0pp), Citation (+14.5pp).
DeepSearchQA — February 2026
Evaluation on Google's DeepSearchQA benchmark: 896 multi-step research questions across 17 fields, independently reproduced by Kaggle.
| Metric | Value |
|---|---|
| Adjusted FC (pass@1) | 84.5% (757 / 896) |
| Automated FC | 83.4% (747 / 896) |
| Avg F1 (concise) | 0.915 |
Grep ranks #1 on the unofficial leaderboard, ahead of Perplexity Deep Research (81.9%), Moonshot K2.5 (77.1%), and the Kaggle-verified #1 Gemini Deep Research Agent (66.1%).
DeepResearch Bench — April 2026
Evaluation on 100 PhD-level research queries using the RACE scoring framework (Gemini-2.5-Pro judge, comparative scoring against expert-written references).
| Dimension | Score |
|---|---|
| Overall | 56.27 |
| Comprehensiveness | 56.79 |
| Insight | 58.98 |
| Instruction Following | 53.49 |
| Readability | 53.50 |
Grep ranks #1 on the leaderboard, ahead of Cellcog Max (56.13), nvidia-aiq (55.95), and Cellcog (55.31).
Parcha Labs Inc — grep.ai