# DROP leaderboard

Reasoning benchmark

> Every figure below is reproduced as its upstream source published it: nothing is modelled, estimated, interpolated or converted. `Unavailable` means no source published the value — it is never a zero. Each value carries its evidence state and the date it was observed.

Page: https://modelchoice.dev/leaderboards/benchmarks/reasoning-drop

Discrete Reasoning Over Paragraphs. Every model the catalog carries a published DROP value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | Reasoning |
| Measure | F1 |
| Tasks | Paragraph reasoning questions |
| Difficulty | Reading and numerical reasoning |
| Published by | [DeepSeek-V4 Technical Report](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf) |

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

## DROP ranking

6 models with a published DROP value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.

| Rank | Model | Provider | F1 | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [o1-mini](https://modelchoice.dev/models/o1-mini) | OpenAI | 83.9 | Observed 2026-11-21 · source benchlm:models |
| 2 | [GPT-4.5](https://modelchoice.dev/models/gpt-4-5) | OpenAI | 83.4 | Observed 2026-11-21 · source benchlm:models |
| 3 | [Claude 3.5 Haiku](https://modelchoice.dev/models/claude-3-5-haiku) | Anthropic | 83.1 | Observed 2026-11-21 · source benchlm:models |
| 4 | [Celeris-1](https://modelchoice.dev/models/celeris-1) | Celeris | 81.4 | Observed 2026-11-21 · source benchlm:models |
| 5 | [o4-mini](https://modelchoice.dev/models/o4-mini) | OpenAI | 77.7 | Observed 2026-11-21 · source benchlm:models |
| 6 | [Soofi S 30B-A3B](https://modelchoice.dev/models/soofi-s-30b-a3b) | Soofi Project | 66.5 | Observed 2026-11-21 · source benchlm:models |

Rows are ordered by the value the source published, highest first. The source does not state whether a higher value is the better result, so this document does not either: for a benchmark that measures a rate of failure, read the table from the bottom.
