# Discrete Reasoning Over Paragraphs (DROP)

> A reading-comprehension benchmark requiring discrete reasoning over paragraphs, reported in DeepSeek-V4 base-model evaluations.

Canonical page: https://benchlm.ai/benchmarks/drop

- Category: [Reasoning](/reasoning)
- Last updated: September 15, 2026

## About DROP

- Year: 2026
- Tasks: Paragraph reasoning questions
- Format: F1
- Difficulty: Reading and numerical reasoning
- Paper: [DeepSeek-V4 Technical Report](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf)

BenchLM stores DROP as a display-only provider-table row when exact values are published in DeepSeek-V4 evaluations.

DROP is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (2 models)

| Rank | Model | Creator | Score |
|------|-------|---------|-------|
| 1 | [Celeris-1](/models/celeris-1) | Celeris | 81.4% |
| 2 | [Soofi S 30B-A3B](/models/soofi-s-30b-a3b) | Soofi Project | 66.5% |

## FAQ

### What does DROP measure?

A reading-comprehension benchmark requiring discrete reasoning over paragraphs, reported in DeepSeek-V4 base-model evaluations.

### Which model scores highest on DROP?

Celeris-1 by Celeris currently leads with a score of 81.4% on DROP.

### How many models are evaluated on DROP?

2 AI models have been evaluated on DROP on BenchLM.

### Does DROP affect BenchLM's overall score?

Not directly. DROP is still displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Compare Top Models on DROP

- [Celeris-1 vs Soofi S 30B-A3B](/compare/celeris-1-vs-soofi-s-30b-a3b)