Intern-S1

👋 加入我们的 Discord 和 微信

简介

我们推出了 Intern-S1-Pro，这是一款万亿参数规模的MoE多模态科学推理模型。Intern-S1-Pro 总参数量达到1T，拥有512个专家，每个token激活8个专家（激活参数为22B）。该模型在高级推理基准测试中表现出色，在AI4Science的关键领域（化学、材料、生命科学、地球科学等）均取得了领先成果，同时保持了强大的通用多模态和文本处理能力。

特点

• 最先进的科学推理能力，在AI4Science任务上与领先的闭源模型不相上下。

• 强大的通用多模态性能，在各类基准测试中表现优异。

• 万亿级MoE训练效率，采用 STE 路由机制（路由网络训练使用密集梯度）和 分组路由 技术，确保稳定收敛并实现专家间的负载均衡。

• 傅里叶位置编码（FoPE）+ 升级的时间序列建模，更好地表示物理信号；支持长且异构的时间序列数据（10^0–10^6个时间点）。

---

Intern-S1简介（点击展开）

我们推出了 Intern-S1，这是我们迄今为止 最先进的开源多模态推理模型。Intern-S1 将 强大的通用任务能力与在广泛科学任务上的顶尖性能 结合在一起，可与领先的闭源商业模型相媲美。

Intern-S1 基于一个235B的MoE语言模型（Qwen3）和一个6B的视觉编码器（InternViT），并在包含超过 2.5万亿个科学领域专用token 的多模态数据上进行了进一步预训练，总数据量达 5万亿token。这使得模型在保持强大通用能力的同时，能够在化学结构解析、蛋白质序列理解以及化合物合成路线规划等专业科学领域表现出色，从而成为现实世界科学研究中的得力助手。

我们还发布了 Intern-S1-mini，这是Intern-S1的轻量级版本，包含一个8B的语言模型和一个0.3B的视觉编码器。

特点

• 在语言和视觉推理基准测试中表现强劲，尤其在科学任务方面尤为突出。

• 持续在海量的5T token数据集上进行预训练，其中超过50%为科学领域专用数据，深度融入了各领域的专业知识。

• 动态分词器能够原生理解分子式、蛋白质序列和地震信号。

模型库

Intern-S1-Pro

	FP8
🤗HuggingFace	internlm/Intern-S1-Pro
ModelScope	Shanghai_AI_Laboratory/Intern-S1-Pro

Intern-S1

	BF16	FP8	GGUF
🤗HuggingFace	internlm/Intern-S1	internlm/Intern-S1-FP8	internlm/Intern-S1-GGUF
ModelScope	Shanghai_AI_Laboratory/Intern-S1	Shanghai_AI_Laboratory/Intern-S1-FP8	Shanghai_AI_Laboratory/Intern-S1-GGUF

Intern-S1-mini

	BF16	FP8	GGUF
🤗HuggingFace	internlm/Intern-S1-mini	internlm/Intern-S1-mini-FP8	internlm/Intern-S1-mini-GGUF
ModelScope	Shanghai_AI_Laboratory/Intern-S1-mini	Shanghai_AI_Laboratory/Intern-S1-mini-FP8	-

性能

我们在包括通用数据集和科学数据集在内的多个基准上对Intern-S1进行了评估。以下是与近期VLM和LLM的性能对比。

Intern-S1-Pro

注：下划线表示开源模型中的最佳性能，加粗表示所有模型中的最佳性能。

Intern-S1

基准	Intern-S1		InternVL3-78B	Qwen2.5-VL-72B	DS-R1-0528	Qwen3-235B-A22B	Kimi-K2-Instruct	Gemini-2.5 Pro	o3	Grok-4
	MMLU-Pro	83.5 ✅		73.0	72.1	83.4	82.2	82.7	86.0	85.0	85.9
MMMU	77.7 ✅		72.2	70.2	-	-	-	81.9	80.8	77.9
GPQA	77.3		49.9	49.0	80.6	71.1	77.8	83.8	83.3	87.5
MMStar	74.9 ✅		72.5	70.8	-	-	-	79.3	75.1	69.6
MathVista	81.5 👑		79.0	74.8	-	-	-	80.3	77.5	72.5
AIME2025	86.0		10.7	10.9	87.5	81.5	51.4	83.0	88.9	91.7
MathVision	62.5 ✅		43.1	38.1	-	-	-	73.0	67.7	67.3
IFEval	86.7		75.6	83.9	79.7	85.0	90.2	91.5	92.2	92.8
SFE	44.3 👑		36.2	30.5	-	-	-	43.0	37.7	31.2
Physics	44.0 ✅		23.1	15.7	-	-	-	40.0	47.9	42.8
SmolInstruct	51.0 👑		19.4	21.0	30.7	28.7	48.1	40.4	43.9	47.3
ChemBench	83.4 👑		61.3	61.6	75.6	75.8	75.3	82.8	81.6	83.3
MatBench	75.0 👑		49.3	51.5	57.7	52.1	61.7	61.7	67.9
MicroVQA	63.9 👑		59.1	53.0	-	-	-	63.1	58.3	59.5
ProteinLMBench	63.1		61.6	61.0	61.4	59.8	66.7	62.9	67.7	66.2
MSEarthMCQ	65.7 👑		57.2	37.6	-	-	-	59.9	61.0	58.0
XLRS-Bench	55.0 👑		49.3	50.9	-	-	-	45.2	43.6	45.4

注：✅表示开源模型中的最佳性能，👑表示所有模型中的最佳性能。

Intern-S1-mini

基准测试	Intern-S1-mini	Qwen3-8B	GLM-4.1V	MiMo-VL-7B-RL-2508
MMLU-Pro	74.78	73.7	57.1	73.93
MMMU	72.33	-	69.9	70.4
MMStar	65.2	-	71.5	72.9
GPQA	65.15	62	50.32	60.35
AIME2024	84.58	76	36.2	72.6
AIME2025	80	67.3	32	64.4
MathVision	51.41	-	53.9	54.5
MathVista	70.3	-	80.7	79.4
IFEval	81.15	85	71.53	71.4
SFE	35.84	-	43.2	43.9
Physics	28.76	-	28.3	28.2
SmolInstruct	32.2	17.6	18.1	16.11
ChemBench	76.47	61.1	56.2	66.78
MatBench	61.55	45.24	54.3	46.9
MicroVQA	56.62	-	50.2	50.96
ProteinLMBench	58.47	59.1	58.3	59.8
MSEarthMCQ	58.12	-	50.3	47.3
XLRS-Bench	51.63	-	49.8	12.29

我们使用 OpenCompass 和 VLMEvalKit 来评估所有模型。 请参考 此页面 以快速开始纯文本评估任务。

用户指南

InternS1 可以使用以下任一 LLM 推理框架进行部署：

• LMDeploy
• vLLM
• SGLang

这些框架的详细部署示例可在以下文档中找到：

• Intern-S1-Pro 模型用户指南
• Intern-S1 和 Intern-S1-Mini 模型用户指南

微调

更多详情请参阅此 文档。

许可证

本项目采用 Apache 2.0 许可证发布。

引用

如果您觉得这项工作有用，请随时引用我们。

@misc{bai2025interns1scientificmultimodalfoundation,
      title={Intern-S1：科学多模态基础模型},
      author={Lei Bai and Zhongrui Cai and Maosong Cao and Weihan Cao and Chiyu Chen and Haojiong Chen and Kai Chen and Pengcheng Chen and Ying Chen and Yongkang Chen and Yu Cheng and Yu Cheng and Pei Chu and Tao Chu and Erfei Cui and Ganqu Cui and Long Cui and Ziyun Cui and Nianchen Deng and Ning Ding and Nanqin Dong and Peijie Dong and Shihan Dou and Sinan Du and Haodong Duan and Caihua Fan and Ben Gao and Changjiang Gao and Jianfei Gao and Songyang Gao and Yang Gao and Zhangwei Gao and Jiaye Ge and Qiming Ge and Lixin Gu and Yuzhe Gu and Aijia Guo and Qipeng Guo and Xu Guo and Conghui He and Junjun He and Yili Hong and Siyuan Hou and Caiyu Hu and Hanglei Hu and Jucheng Hu and Ming Hu and Zhouqi Hua and Haian Huang and Junhao Huang and Xu Huang and Zixian Huang and Zhe Jiang and Lingkai Kong and Linyang Li and Peiji Li and Pengze Li and Shuaibin Li and Tianbin Li and Wei Li and Yuqiang Li and Dahua Lin and Junyao Lin and Tianyi Lin and Zhishan Lin and Hongwei Liu and Jiangning Liu and Jiyao Liu and Junnan Liu and Kai Liu and Kaiwen Liu and Kuikun Liu and Shichun Liu and Shudong Liu and Wei Liu and Xinyao Liu and Yuhong Liu and Zhan Liu and Yinquan Lu and Haijun Lv and Hongxia Lv and Huijie Lv and Qidang Lv and Ying Lv and Chengqi Lyu and Chenglong Ma and Jianpeng Ma and Ren Ma and Runmin Ma and Runyuan Ma and Xinzhu Ma and Yichuan Ma and Zihan Ma and Sixuan Mi and Junzhi Ning and Wenchang Ning and Xinle Pang and Jiahui Peng and Runyu Peng and Yu Qiao and Jiantao Qiu and Xiaoye Qu and Yuan Qu and Yuchen Ren and Fukai Shang and Wenqi Shao and Junhao Shen and Shuaike Shen and Chunfeng Song and Demin Song and Diping Song and Chenlin Su and Weijie Su and Weigao Sun and Yu Sun and Qian Tan and Cheng Tang and Huanze Tang and Kexian Tang and Shixiang Tang and Jian Tong and Aoran Wang and Bin Wang and Dong Wang and Lintao Wang and Rui Wang and Weiyun Wang and Wenhai Wang and Yi Wang and Ziyi Wang and Ling-I Wu and Wen Wu and Yue Wu and Zijian Wu and Linchen Xiao and Shuhao Xing and Chao Xu and Huihui Xu and Jun Xu and Ruiliang Xu and Wanghan Xu and GanLin Yang and Yuming Yang and Haochen Ye and Jin Ye and Shenglong Ye and Jia Yu and Jiashuo Yu and Jing Yu and Fei Yuan and Bo Zhang and Chao Zhang and Chen Zhang and Hongjie Zhang and Jin Zhang and Qiaosheng Zhang and Qiuyinzhe Zhang and Songyang Zhang and Taolin Zhang and Wenlong Zhang and Wenwei Zhang and Yechen Zhang and Ziyang Zhang and Haiteng Zhao and Qian Zhao and Xiangyu Zhao and Bowen Zhou and Dongzhan Zhou and Peiheng Zhou and Yuhao Zhou and Yunhua Zhou and Dongsheng Zhu and Lin Zhu and Yicheng Zou},
      year={2025},
      eprint={2508.15763},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2508.15763},
}

@misc{zou2026interns1proscientificmultimodalfoundation,
      title={Intern-S1-Pro：万亿规模的科学多模态基础模型}, 
      author={Yicheng Zou and Dongsheng Zhu and Lin Zhu and Tong Zhu and Yunhua Zhou and Peiheng Zhou and Xinyu Zhou and Dongzhan Zhou and Zhiwang Zhou and Yuhao Zhou and Bowen Zhou and Zhanping Zhong and Zhijie Zhong and Haiteng Zhao and Penghao Zhao and Xiaomeng Zhao and Zhiyuan Zhao and Yechen Zhang and Jin Zhang and Wenwei Zhang and Hongjie Zhang and Zhuo Zhang and Wenlong Zhang and Bo Zhang and Chao Zhang and Chen Zhang and Yuhang Zang and Fei Yuan and Jiakang Yuan and Jiashuo Yu and Jinhui Yin and Haochen Ye and Qian Yao and Bowen Yang and Danni Yang and Kaichen Yang and Ziang Yan and Jun Xu and Yicheng Xu and Wanghan Xu and Xuenan Xu and Chao Xu and Ruiliang Xu and Shuhao Xing and Long Xing and Xinchen Xie and Ling-I Wu and Zijian Wu and Zhenyu Wu and Lijun Wu and Yue Wu and Jianyu Wu and Wen Wu and Fan Wu and Xilin Wei and Qi Wei and Bingli Wang and Rui Wang and Ziyi Wang and Zun Wang and Yi Wang and Haomin Wang and Yizhou Wang and Lintao Wang and Yiheng Wang and Longjiang Wang and Bin Wang and Jian Tong and Zhongbo Tian and Huanze Tang and Chen Tang and Shixiang Tang and Yu Sun and Qiushi Sun and Xuerui Su and Qisheng Su and Chenlin Su and Demin Song and Jin Shi and Fukai Shang and Yuchen Ren and Pengli Ren and Xiaoye Qu and Yuan Qu and Jiantao Qiu and Yu Qiao and Runyu Peng and Tianshuo Peng and Jiahui Peng and Qizhi Pei and Zhuoshi Pan and Linke Ouyang and Wenchang Ning and Yichuan Ma and Zerun Ma and Ningsheng Ma and Runyuan Ma and Chengqi Lyu and Haijun Lv and Han Lv and Lindong Lu and Kuikun Liu and Jiangning Liu and Yuhong Liu and Kai Liu and Hongwei Liu and Zhoumianze Liu and Mengjie Liu and Ziyu Liu and Wenran Liu and Yang Liu and Liwei Liu and Kaiwen Liu and Junyao Lin and Junming Lin and Tianyang Lin and Dahua Lin and Jianze Liang and Linyang Li and Peiji Li and Zonglin Li and Zehao Li and Pengze Li and Guoyan Li and Lingkai Kong and Linglin Jing and Zhenjiang Jin and Feifei Jiang and Qian Jiang and Junhao Huang and Zixian Huang and Haian Huang and Zhouqi Hua and Han Hu and Linfeng Hou and Yinan He and Conghui He and Tianyao He and Xu Guo and Qipeng Guo and Aijia Guo and Yuzhe Gu and Lixin Gu and Jingyang Gong and Qiming Ge and Jiaye Ge and Songyang Gao and Jianfei Gao and Xinyu Fang and Caihua fan and Yue Fan and Yanhui Duan and Zichen Ding and Shengyuan Ding and Xuanlang Dai and Erfei Cui and Ganqu Cui and Pei Chu and Tao Chu and Guangran Cheng and Yu Cheng and Kai Chen and Yongkang Chen and Chiyu Chen and Guanzhou Chen and Qiaosheng Chen and Sitao Chen and Xin Chen and Haojiong Chen and Yicheng Chen and Weihan Cao and Yuhang Cao and Qinglong Cao and Lei Bai},
      year={2026},
      eprint={2603.25040},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2603.25040}, 
}

Intern-S1 快速上手指南

Intern-S1 是上海人工智能实验室推出的开源多模态科学推理模型系列，包含万亿参数 MoE 架构的 Intern-S1-Pro、高性能的 Intern-S1 以及轻量级的 Intern-S1-mini。该模型在化学、材料、生命科学及地球科学等 AI4Science 领域表现卓越，同时具备强大的通用多模态能力。

环境准备

系统要求

• 操作系统: Linux (推荐 Ubuntu 20.04+)
• GPU: NVIDIA GPU (推荐 A100/H100 或同等算力显存)
  • Intern-S1-Pro: 建议多卡部署或使用 FP8 量化版本以降低显存需求。
  • Intern-S1 / Mini: 单卡或多卡均可，视具体量化版本而定。
• CUDA: 12.1 或更高版本
• Python: 3.10+

前置依赖

确保已安装以下基础库：

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
pip install transformers>=4.45.0 accelerate sentencepiece protobuf

提示：国内用户推荐使用清华或阿里镜像源加速安装：

pip install -i https://pypi.tuna.tsinghua.edu.cn/simple torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu121

安装步骤

方式一：通过 Hugging Face Transformers 直接加载（推荐）

无需克隆仓库，直接使用 transformers 库加载模型。

pip install -i https://pypi.tuna.tsinghua.edu.cn/simple transformers accelerate torch

方式二：从 ModelScope 下载（国内加速）

对于国内开发者，使用 ModelScope 可获得更快的下载速度。

pip install -i https://pypi.tuna.tsinghua.edu.cn/simple modelscope

基本使用

示例 1：使用 Hugging Face 加载 Intern-S1-mini 进行推理

以下是最简单的文本/多模态推理示例。

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# 配置模型路径 (可选用 internlm/Intern-S1-mini, internlm/Intern-S1, 或 internlm/Intern-S1-Pro)
model_path = "internlm/Intern-S1-mini" 

# 加载分词器和模型 (自动下载)
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_path, 
    device_map="auto", 
    trust_remote_code=True,
    torch_dtype=torch.bfloat16 # 建议使用 bfloat16 或 float16
)

# 构建输入提示 (科学推理示例)
prompt = "请解释蛋白质序列中疏水氨基酸的作用，并给出一个具体的例子。"
messages = [
    {"role": "user", "content": prompt}
]

# 生成回复
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer([text], return_tensors="pt").to(model.device)

outputs = model.generate(
    **inputs, 
    max_new_tokens=512, 
    do_sample=True, 
    temperature=0.7, 
    top_p=0.8
)

response = tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True)
print(response)

示例 2：使用 ModelScope 加载（国内网络优化）

如果你在中国大陆，使用 ModelScope 加载模型通常更稳定。

from modelscope import AutoModelForCausalLM, AutoTokenizer
import torch

# 使用 ModelScope 的模型 ID
model_path = "Shanghai_AI_Laboratory/Intern-S1-mini"

tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_path, 
    device_map="auto", 
    trust_remote_code=True,
    torch_dtype=torch.bfloat16
)

prompt = "分析这张化学结构图可能代表的化合物性质。（需配合图像输入处理代码）"
# 注意：多模态输入需要按照特定格式处理图像 tensor，此处仅展示文本调用框架
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

显存优化建议

• FP8 版本：如果显存受限，请优先选择 -FP8 结尾的模型版本（如 internlm/Intern-S1-FP8），需安装 bitsandbytes 或相关 FP8 支持库。
• GGUF 版本：如需在 CPU 或消费级显卡运行，可下载 GGUF 版本并使用 llama.cpp 或 ollama 运行。