Datasets:

Modalities:
Image
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
pandas
License:
RxnBench / README.md
AI4Industry's picture
Add results of Gemini-3-Flash-Preview and Seed-1.8
7bfe2c8 verified
---
dataset_info:
features:
- name: image
dtype: image
- name: question
dtype: string
- name: choices
list: string
- name: answer
dtype: int32
- name: meta_info
struct:
- name: title
dtype: string
- name: journal
dtype: string
- name: doi
dtype: string
- name: url
dtype: string
- name: question_type
dtype: string
splits:
- name: en
num_bytes: 546653187.125
num_examples: 1525
- name: zh
num_bytes: 546319847.125
num_examples: 1525
download_size: 218606009
dataset_size: 1092973034.25
configs:
- config_name: RxnBench-VQA
data_files:
- split: en
path: data/en-*
- split: zh
path: data/zh-*
license: cc-by-nc-sa-4.0
task_categories:
- visual-question-answering
language:
- en
- zh
tags:
- chemistry
---
# RxnBench: A Benchmark for Chemical Reaction Figure Understanding
## ๐Ÿ“˜ Benchmark Summary
RxnBench (SF-QA) is a visual question answering (VQA) benchmark comprising 1,525 multiple-choice questions (MCQs) at the PhD-level of organic chemistry reaction understanding.
The benchmark is built from 305 scientific figures drawn from high-impact OpenAssess journals.
For each figure, domain experts carefully designed five multiple-choice VQA questions targeting the interpretation of organic reaction diagrams.
These questions were further refined through multiple rounds of rigorous review and revision to ensure both clarity and scientific accuracy.
The questions cover a variety of types, including the description of chemical reaction images, extraction of reaction content, recognition of molecules or Markush structures, and determination of mechanisms.
This benchmark challenges visual-language models on their foundational knowledge of organic chemistry, multimodal contextual reasoning, and chemical reasoning skills.
The benchmark is released in both English and Chinese versions.
## ๐Ÿ“‘ Task Types
We categorize chemical reaction visual question answering tasks into six types:
- **Type 0 โ€” Fact Extraction**: Direct retrieval of textual or numerical information from reaction schemes.
- **Type 1 โ€” Reagent Roles and Functions Identification**: Identification of reagents and their functional roles, requiring chemical knowledge and reaction-type awareness.
- **Type 2 โ€” Reaction Mechanism and Process Understanding**: Interpretation of reaction progression, including intermediates, catalytic cycles, and mechanistic steps.
- **Type 3 โ€” Comparative Analysis and Reasoning**: Comparative evaluation, causal explanation, or outcome prediction under varying conditions.
- **Type 4 โ€” Multi-step Synthesis and Global Understanding**: Comprehension of multi-step pathways, step-to-step coherence, and overall synthetic design.
- **Type 5 โ€” Chemical Structure Recognition**: Extraction and reasoning-based parsing of chemical structures in SMILES or E-SMILES (as defined in the [MolParser](https://arxiv.org/abs/2411.11098) paper).
![output3](https://cdn-uploads.huggingface.co/production/uploads/65f7f16fb6941db5c2e7c4bf/oTOMcZE7oz-Pv4fUUpi0J.png)
## ๐ŸŽฏ Benchmark Evaluation
This benchmark evaluates model performance on multiple-choice question answering (MCQ) tasks.
We provide two versions of the prompt template, depending on the language setting.
**English Prompt**
```
Question: {question}
Choices:
A. {choice_A}
B. {choice_B}
C. {choice_C}
D. {choice_D}
Based on the image and the question, choose the most appropriate answer.
**Only output a single letter (A, B, C, or D)**. Do NOT output any other text or explanation.
```
**Chinese Prompt**
```
้—ฎ้ข˜: {question}
้€‰้กน:
A. {choice_A}
B. {choice_B}
C. {choice_C}
D. {choice_D}
่ฏทๆ นๆฎๅ›พๅƒๅ’Œ้—ฎ้ข˜๏ผŒไปŽไปฅไธŠๅ››ไธช้€‰้กนไธญ้€‰ๆ‹ฉๆœ€ๅˆ้€‚็š„็ญ”ๆกˆใ€‚
ๅช่พ“ๅ‡บๅ•ไธชๅญ—ๆฏ (A, B, C ๆˆ– D)๏ผŒไธ่ฆ่พ“ๅ‡บ้€‰้กนๅ†…ๅฎน๏ผŒไนŸไธ่ฆ่พ“ๅ‡บไปปไฝ•่งฃ้‡Šใ€‚
```
**Evaluation Protocol**
If the modelโ€™s output is not one of A, B, C, or D, we use GPT-4o to map the output to Aโ€“D based on the option content.
The final evaluation reports the absolute accuracy of the benchmark in both English and Chinese versions.
Code Example: https://github.com/uni-parser/RxnBench
## ๐Ÿ“Š Benchmark Leaderboard
We evaluated several of the latest popular MLLMs, including both closed-source and open-source models.
| Moldel | Think | Weight | API-Version | RxnBench-En | RxnBench-Zh | Mean Score |
| ---- |:----:|:----:|:----:|:----:|:----:|:----:|
| Gemini-3-Flash-preview | โˆš | Proprietary | 20251217 | **0.9593** | **0.9652** | **0.9623** |
| Seed1.8-Think | โˆš | Proprietary | 20251218 | 0.9325 | 0.9403 | 0.9364 |
| Gemini-3-Pro-preview | โˆš | Proprietary | 20251119 | 0.9318 | 0.9403 | 0.9361 |
| GPT-5(high) | โˆš | Proprietary | 20250807 | 0.9279 | 0.9246 | 0.9263 |
| Gemini-2.5-Pro | โˆš | Proprietary | 20250617 | 0.9095 | 0.9423 | 0.9259 |
| GPT-5.1(high) | โˆš | Proprietary | 20251113 | 0.9213 | 0.9220 | 0.9216 |
| GPT-5(medium) | โˆš | Proprietary | 20250807 | 0.9207 | 0.9226 | 0.9216 |
| Qwen3-VL-235BA22B-Think | โˆš | Open | - | 0.9220 | 0.9134 | 0.9177 |
| Qwen3-VL-32B-Think | โˆš | Open | - | 0.9128 | 0.9161 | 0.9144 |
| GPT-5.1(medium) | โˆš | Proprietary | 20251113 | 0.9108 | 0.9141 | 0.9125 |
| GPT-5-mini | โˆš | Proprietary | 20250807 | 0.9108 | 0.9128 | 0.9118 |
| Seed1.5-VL-Think | โˆš | Proprietary | 20250428 | 0.9056 | 0.9161 | 0.9109 |
| GPT o3 | โˆš | Proprietary | 20250416 | 0.9056 | 0.9115 | 0.9086 |
| GPT o4 mini | โˆš | Proprietary | 20250416 | 0.9062 | 0.9075 | 0.9069 |
| InternVL3.5-241B-A28B | โˆš | Open | - | 0.9003 | 0.9062 | 0.9033 |
| Intern-S1 | โˆš | Open | - | 0.8938 | 0.8944 | 0.8941 |
| Qwen3-VL-30BA3B-Think | โˆš | Open | - | 0.8689 | 0.8590 | 0.8689 |
| Qwen3-VL-Plus | ร— | Proprietary | 20250923 | 0.8551 | 0.8656 | 0.8604 |
| Qwen3-VL-8B-Think | โˆš | Open | - | 0.8636 | 0.8564 | 0.8600 |
| Seed1.5-VL | ร— | Proprietary | 20250328 | 0.8518 | 0.8669 | 0.8594 |
| Qwen3-VL-235BA22B-Instruct | ร— | Open | - | 0.8492 | 0.8675 | 0.8584 |
| InternVL3-78b | ร— | Open | - | 0.8531 | 0.8308 | 0.8420 |
| Qwen3-VL-4B-Think | โˆš | Open | - | 0.8577 | 0.8256 | 0.8416 |
| Intern-S1-mini | โˆš | Open | - | 0.8521 | 0.8282 | 0.8402 |
| GLM-4.1V-9B-Thinking | โˆš | Open | - | 0.8392 | 0.8341 | 0.8367 |
| Qwen3-VL-32B-Instruct | ร— | Open | - | 0.8315 | 0.8407 | 0.8361 |
| Qwen2.5-VL-72B | ร— | Open | - | 0.8341 | 0.8308 | 0.8325 |
| Qwen2.5-VL-Max | ร— | Proprietary | 20250813 | 0.8192 | 0.8262 | 0.8227 |
| GPT-5-nano | โˆš | Proprietary | 20250807 | 0.7980 | 0.7941 | 0.7961 |
| Qwen2.5-VL-32B | ร— | Open | - | 0.7980 | 0.7908 | 0.7944 |
| Gemini-2.5-Flash | โˆš | Proprietary | 20250617 | 0.6925 | 0.8557 | 0.7741 |
| Qwen3-VL-8B-Instruct | ร— | Open | - | 0.7548 | 0.7495 | 0.7521 |
| Qwen3-VL-30BA3B-Instruct | ร— | Open | - | 0.7456 | 0.7436 | 0.7456 |
| GPT-4o | ร— | Proprietary | 20240806 | 0.7462 | 0.7436 | 0.7449 |
| Qwen2.5-VL-7B | ร— | Open | - | 0.7082 | 0.7233 | 0.7158 |
| Qwen3-VL-4B-Instruct | ร— | Open | - | 0.7023 | 0.7023 | 0.7023 |
| Qwen3-VL-2B-Think | โˆš | Open | - | 0.6780 | 0.6708 | 0.6744 |
| Qwen2.5-VL-3B | ร— | Open | - | 0.6748 | 0.6643 | 0.6696 |
| GPT-4o mini | ร— | Proprietary | 20240718 | 0.6636 | 0.6066 | 0.6351 |
| Qwen3-VL-2B-Instruct | ร— | Open | - | 0.5711 | 0.5928 | 0.5820 |
| *Choice longest answer* | - | - | - | 0.4262 | 0.4525 | 0.4394 |
| Deepseek-VL2 | ร— | Open | - | 0.4426 | 0.4216 | 0.4321 |
| *Random* | - | - | - | 0.2500 | 0.2500 | 0.2500 |
We also conducted separate evaluations for different task types (in RxnBench-en).
| Moldel | Think | Weight | API-Version | Type0 | Type1 | Type2 | Type3 | Type4 | Type5 |
| ---- |:----:|:----:|:----:|:----:|:----:|:----:|:----:|:----:|:----:|
| Gemini-3-Flash-preview | โˆš | Proprietary | 20251217 | 0.9613 | **0.9643** | **0.9764** | **0.9630** | 0.9492 | **0.9030** |
| Seed1.8-Think | โˆš | Proprietary | 20251218 | 0.9331 | 0.9484 | 0.9527 | 0.9444 | 0.9492 | 0.8284 |
| Gemini-3-Pro-preview | โˆš | Proprietary | 20251119 | **0.9648** | 0.9246 | 0.9527 | 0.9398 | 0.9322 | 0.7463 |
| GPT-5(high) | โˆš | Proprietary | 20250807 | 0.9313 | 0.9444 | 0.9527 | 0.9167 | **0.9661** | 0.8358 |
| Gemini-2.5-Pro | โˆš | Proprietary | 20250617 | 0.9331 | 0.9246 | 0.9459 | 0.9491 | 0.9322 | 0.6343 |
| GPT-5.1(high) | โˆš | Proprietary | 20251113 | 0.9243 | 0.9524 | 0.9426 | 0.9167 | 0.9661 | 0.7910 |
| GPT-5(medium) | โˆš | Proprietary | 20250807 | 0.9349 | 0.9325 | 0.9493 | 0.9167 | 0.9492 | 0.7761 |
| Qwen3-VL-235BA22B-Think | โˆš | Open | - | 0.9190 | 0.9405 | 0.9459 | 0.9213 | 0.9322 | 0.8433 |
| Qwen3-VL-32B-Think | โˆš | Open | - | 0.9296 | 0.9405 | 0.9426 | 0.9259 | 0.9153 | 0.7015 |
| GPT-5.1(medium) | โˆš | Proprietary | 20251113 | 0.9243 | 0.9365 | 0.9426 | 0.9167 | 0.9492 | 0.7090 |
| GPT-5-mini | โˆš | Proprietary | 20250807 | 0.9225 | 0.9325 | 0.9257 | 0.9259 | 0.9831 | 0.7388 |
| Seed1.5-VL-Think | โˆš | Proprietary | 20250428 | 0.8996 | 0.9365 | 0.9358 | 0.9074 | 0.9153 | 0.8060 |
| GPT o3 | โˆš | Proprietary | 20250416 | 0.9313 | 0.9325 | 0.9223 | 0.8981 | 0.9492 | 0.7090 |
| GPT o4 mini | โˆš | Proprietary | 20250416 | 0.6391 | 0.7302 | 0.7500 | 0.6667 | 0.6271 | 0.4627 |
| InternVL3.5-241B-A28B | โˆš | Open | - | 0.8944 | 0.9127 | 0.9291 | 0.9167 | 0.9153 | 0.8134 |
| Intern-S1 | โˆš | Open | - | 0.9014 | 0.9127 | 0.9223 | 0.9028 | 0.8814 | 0.7463 |
| Qwen3-VL-30BA3B-Think | โˆš | Open | - | 0.8732 | 0.8810 | 0.9054 | 0.8843 | 0.9322 | 0.6940 |
| Qwen3-VL-Plus | ร— | Proprietary | 20250923 | 0.8275 | 0.8968 | 0.8986 | 0.8565 | 0.9153 | 0.7687 |
| Qwen3-VL-8B-Think | โˆš | Open | - | 0.8768 | 0.8730 | 0.8885 | 0.9028 | 0.8983 | 0.6567 |
| Seed1.5-VL | ร— | Proprietary | 20250328 | 0.9327 | 0.9127 | 0.9122 | 0.8472 | 0.8305 | 0.7015 |
| Qwen3-VL-235BA22B-Instruct | ร— | Open | - | 0.8204 | 0.8929 | 0.8986 | 0.8426 | 0.8814 | 0.7761 |
| InternVL3-78b | ร— | Open | - | 0.8556 | 0.8730 | 0.8885 | 0.8981 | 0.9153 | 0.6194 |
| Qwen3-VL-4B-Think | โˆš | Open | - | 0.8838 | 0.8770 | 0.8615 | 0.9074 | 0.8983 | 0.6045 |
| Intern-S1-mini | โˆš | Open | - | 0.8239 | 0.8690 | 0.8547 | 0.8611 | 0.8475 | 0.6791 |
| GLM-4.1V-9B-Thinking | โˆš | Open | - | 0.8433 | 0.8690 | 0.8649 | 0.8657 | 0.8814 | 0.6493 |
| Qwen3-VL-32B-Instruct | ร— | Open | - | 0.8169 | 0.8571 | 0.8885 | 0.8519 | 0.8305 | 0.6866 |
| Qwen2.5-VL-72B | ร— | Open | - | 0.8063 | 0.8063 | 0.8770 | 0.9088 | 0.8102 | 0.9322 | 0.7090 |
| Qwen2.5-VL-Max | ร— | Proprietary | 20250813 | 0.7958 | 0.8571 | 0.8885 | 0.8194 | 0.8983 | 0.6642 |
| GPT-5-nano | โˆš | Proprietary | 20250807 | 0.8063 | 0.8452 | 0.8311 | 0.8241 | 0.7797 | 0.5672 |
| Qwen2.5-VL-32B | ร— | Open | - | 0.7729 | 0.8413 | 0.8750 | 0.8009 | 0.8305 | 0.6418 |
| Gemini-2.5-Flash | โˆš | Proprietary | 20250617 | 0.7799 | 0.6111 | 0.6757 | 0.6620 | 0.7627 | 0.5373 |
| Qwen3-VL-8B-Instruct | ร— | Open | - | 0.7113 | 0.8175 | 0.8446 | 0.8241 | 0.7627 | 0.5075 |
| Qwen3-VL-30BA3B-Instruct | ร— | Open | - | 0.7042 | 0.7937 | 0.8311 | 0.7824 | 0.7119 | 0.5970 |
| GPT-4o | ร— | Proprietary | 20240806 | 0.7359 | 0.8175 | 0.7973 | 0.7500 | 0.7627 | 0.5224 |
| Qwen2.5-VL-7B | ร— | Open | - | 0.6678 | 0.7659 | 0.8041 | 0.7130 | 0.6441 | 0.5373 |
| Qwen3-VL-4B-Instruct | ร— | Open | - | 0.6708 | 0.7302 | 0.7804 | 0.7222 | 0.6610 | 0.5970 |
| Qwen3-VL-2B-Think | โˆš | Open | - | 0.7342 | 0.6706 | 0.7128 | 0.7083 | 0.6102 | 0.3657 |
| Qwen2.5-VL-3B | ร— | Open | - | 0.6426 | 0.7381 | 0.7635 | 0.6898 | 0.6610 | 0.4776 |
| GPT-4o mini | ร— | Proprietary | 20240718 | 0.6391 | 0.7302 | 0.7500 | 0.6667 | 0.6271 | 0.4627 |
| Qwen3-VL-2B-Instruct | ร— | Open | - | 0.5405 | 0.6190 | 0.6318 | 0.6250 | 0.6102 | 0.3731 |
| Deepseek-VL2 | ร— | Open | - | 0.4120 | 0.5040 | 0.4899 | 0.4907 | 0.3729 | 0.3060 |
## ๐Ÿ†• RxnBench-Doc
A single reaction image often lacks the information needed for full interpretation, requiring contextual text from the literature. Therefore, we also provide a benchmark for chemical reaction literature understanding.
https://huggingface.co/datasets/UniParser/RxnBench-Doc
## ๐Ÿ“– Citation
our paper coming soon ...