Hugging Face's logo

stepfun-ai
/
step3-fp8

Image-Text-to-Text
Transformers
step3_vl
text-generation
conversational
custom_code
fp8
Model card Files
xet
Community
4
step3-fp8
File size: 11,428 Bytes 
565af60
 
098d4ad
 
565af60
098d4ad
cc0b4a3
 
48df371
cc0b4a3
 
 
 
 
 
 
 
 
 
 
adcb82c
565af60
cc0b4a3
 
 
 
 
565af60
cc0b4a3
 
 
098d4ad
cc0b4a3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b197bac
 
cc0b4a3
 
 
6f7bdf5
cc0b4a3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
adcb82c
cc0b4a3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cca32c9
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
 
---
library_name: transformers
license: apache-2.0
pipeline_tag: image-text-to-text
---

<div align="center">
  <picture>
      <img src="https://github.com/stepfun-ai/Step3/blob/main/figures/stepfun-logo.png?raw=true" width="30%" alt="StepFun: Cost-Effective Multimodal Intelligence">
  </picture>
</div>

<hr>

<div align="center" style="line-height:1">
  <a href="https://stepfun.com/" target="_blank"><img alt="Chat" src="https://img.shields.io/badge/Chat-StepFun-ff6b6b?color=1783ff&logoColor=white"/></a>
  <a href="https://stepfun.com/" target="_blank"><img alt="Homepage" src="https://img.shields.io/badge/Homepage-StepFun-white?logo=StepFun&logoColor=white"/></a>
</div>

<div align="center" style="line-height: 1;">
  <a href="https://github.com/stepfun-ai/Step3" target="_blank"><img alt="GitHub" src="https://img.shields.io/badge/GitHub-StepFun-white?logo=github&logoColor=white"/></a>
  <a href="https://www.modelscope.cn/models/stepfun-ai/step3" target="_blank"><img alt="ModelScope" src="https://img.shields.io/badge/🤖ModelScope-StepFun-ffc107?color=7963eb&logoColor=white"/></a>
  <a href="https://x.com/StepFun_ai" target="_blank"><img alt="Twitter Follow" src="https://img.shields.io/badge/Twitter-StepFun-white?logo=x&logoColor=white"/></a>
</div>

<div align="center" style="line-height: 1;">
<a href="https://discord.com/invite/XHheP5Fn" target="_blank"><img alt="Discord" src="https://img.shields.io/badge/Discord-StepFun-white?logo=discord&logoColor=white"/></a>
  <a href="https://huggingface.co/stepfun-ai/step3-fp8/blob/main/LICENSE"><img alt="License" src="https://img.shields.io/badge/License-Apache%202.0-blue?&color=blue"/></a>
</div>

<div align="center">
<b>📰&nbsp;&nbsp;<a href="https://stepfun.ai/research/step3">Step3 Model Blog</a></b> &nbsp;&nbsp;&nbsp; | &nbsp;&nbsp;&nbsp; <b>📄&nbsp;&nbsp;<a href="https://huggingface.co/papers/2507.19427">Paper</a></b>
</div>

## Introduction

Step3 is our cutting-edge multimodal reasoning model—built on a Mixture-of-Experts architecture with 321B total parameters and 38B active. 
It is designed end-to-end to minimize decoding costs while delivering top-tier performance in vision–language reasoning. 
Through the co-design of Multi-Matrix Factorization Attention (MFA) and Attention-FFN Disaggregation (AFD), 
Step3 maintains exceptional efficiency across both flagship and low-end accelerators.

### Step3 model card:

|          Config        |  Value  |
|------------------------|---------|
| **Number of Layers (Dense layer included)**|61|
|**Number of Dense Layers**| 5|
| **Hidden Dimension**       | 7168    |
| **Attention Mechanism**    | MFA     |
| **Low-rank Query Dimension** | 2048  |
| **Number of Query Heads**          | 64      |
| **Head Dimension**        | 256     |
|**Number of Experts** |48|
|**Selected Experts per Token**|3|
|**Number of Shared Experts**| 1|
| **Max Context Length** | 65536 |
| **Tokenizer** | Deepseek V3 |
| **Total Parameters (LLM)** | 316B |
| **Activated Params per Token** | 38B |
| **Total Parameters (VLM)** | 321B |


## Evaluation Results
<table>
  <thead>
    <tr>
      <th></th>
      <th>Model</th>
      <th>Total Params.</th>
      <th>MMMU</th>
      <th>MathVision</th>
      <th>ZeroBench(sub)</th>
      <th>DYNAMATH</th>
      <th>SimpleVQA</th>
      <th>HallusionBench</th>
      <th>AIME25</th>
      <th>HMMT25</th>
      <th>CNMO24</th>
      <th>GPQA-Diamond</th>
      <th>LiveCodeBench<br>(24.8-25.5)</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td rowspan="6">Open-Source VLM</td>
      <td>Step3</td>
      <td>321B</td>
      <td>74.2</td>
      <td>64.8</td>
      <td>23.0</td>
      <td>50.1</td>
      <td>62.2</td>
      <td>64.2</td>
      <td>82.9</td>
      <td>70.0</td>
      <td>83.7</td>
      <td>73.0</td>
      <td>67.1</td>
    </tr>
    <tr>
      <td>ERINE4.5 - thinking</td>
      <td>300B/424B</td>
      <td>70.0</td>
      <td>47.6</td>
      <td>22.5</td>
      <td>46.9</td>
      <td>59.8</td>
      <td>60.0</td>
      <td>35.1</td>
      <td>40.5*</td>
      <td>75.5</td>
      <td>76.8</td>
      <td>38.8</td>
    </tr>
    <tr>
      <td>GLM-4.1V-thinking</td>
      <td>9B</td>
      <td>68.0</td>
      <td>49.4</td>
      <td>22.8</td>
      <td>41.9</td>
      <td>48.1</td>
      <td>60.8</td>
      <td>13.3</td>
      <td>6.7</td>
      <td>25.0</td>
      <td>47.4</td>
      <td>24.2</td>
    </tr>
    <tr>
      <td>MiMo-VL</td>
      <td>7B</td>
      <td>66.7</td>
      <td>60.4</td>
      <td>18.6</td>
      <td>45.9</td>
      <td>48.5</td>
      <td>59.6</td>
      <td>60.0</td>
      <td>34.6</td>
      <td>69.9</td>
      <td>55.5</td>
      <td>50.1</td>
    </tr>
    <tr>
      <td>QvQ-72B-Preview</td>
      <td>72B</td>
      <td>70.3</td>
      <td>35.9</td>
      <td>15.9</td>
      <td>30.7</td>
      <td>40.3</td>
      <td>50.8</td>
      <td>22.7</td>
      <td>49.5</td>
      <td>47.3</td>
      <td>10.9</td>
      <td>24.1</td>
    </tr>
    <tr>
      <td>LLaMA-Maverick</td>
      <td>400B</td>
      <td>73.4</td>
      <td>47.2</td>
      <td>22.8</td>
      <td>47.1</td>
      <td>45.4</td>
      <td>57.1</td>
      <td>19.2</td>
      <td>8.91</td>
      <td>41.6</td>
      <td>69.8</td>
      <td>33.9</td>
    </tr>
    <tr>
      <td rowspan="4">Open-Source LLM</td>
      <td>MiniMax-M1-80k</td>
      <td>456B</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>76.9</td>
      <td>-</td>
      <td>-</td>
      <td>70.0</td>
      <td>65.0</td>
    </tr>
    <tr>
      <td>Qwen3-235B-A22B-Thinking</td>
      <td>235B</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>81.5</td>
      <td>62.5</td>
      <td>-</td>
      <td>71.1</td>
      <td>65.9</td>
    </tr>
    <tr>
      <td>DeepSeek R1-0528</td>
      <td>671B</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>87.5</td>
      <td>79.4</td>
      <td>86.9</td>
      <td>81.0</td>
      <td>73.3</td>
    </tr>
    <tr>
      <td>Qwen3-235B-A22B-Thinking-2507</td>
      <td>235B</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>-</td>
      <td>92.3</td>
      <td>83.9</td>
      <td>-</td>
      <td>81.1</td>
      <td>-</td>
    </tr>
    <tr>
      <td rowspan="6">Proprietary VLM</td>
      <td>O3</td>
      <td>-</td>
      <td>82.9</td>
      <td>72.8</td>
      <td>25.2</td>
      <td>58.1</td>
      <td>59.8</td>
      <td>60.1</td>
      <td>88.9</td>
      <td>70.1</td>
      <td>86.7</td>
      <td>83.3</td>
      <td>75.8</td>
    </tr>
    <tr>
      <td>Claude4 Sonnet (thinking)</td>
      <td>-</td>
      <td>76.9</td>
      <td>64.6</td>
      <td>26.1</td>
      <td>48.1</td>
      <td>43.7</td>
      <td>57.0</td>
      <td>70.5</td>
      <td>-</td>
      <td>-</td>
      <td>75.4</td>
      <td>55.9</td>
    </tr>
    <tr>
      <td>Claude4 opus (thinking)</td>
      <td>-</td>
      <td>79.8</td>
      <td>66.1</td>
      <td>25.2</td>
      <td>49.3</td>
      <td>47.2</td>
      <td>59.9</td>
      <td>75.5</td>
      <td>-</td>
      <td>-</td>
      <td>79.6</td>
      <td>56.6</td>
    </tr>
    <tr>
      <td>Gemini 2.5 Flash (thinking)</td>
      <td>-</td>
      <td>73.2</td>
      <td>57.3</td>
      <td>20.1</td>
      <td>57.1</td>
      <td>61.1</td>
      <td>65.2</td>
      <td>72.0</td>
      <td>-</td>
      <td>-</td>
      <td>82.8</td>
      <td>61.9</td>
    </tr>
    <tr>
      <td>Gemini 2.5 Pro</td>
      <td>-</td>
      <td>81.7</td>
      <td>73.3</td>
      <td>30.8</td>
      <td>56.3</td>
      <td>66.8</td>
      <td>66.8</td>
      <td>88.0</td>
      <td>-</td>
      <td>-</td>
      <td>86.4</td>
      <td>71.8</td>
    </tr>
    <!-- 新增 Grok 4 -->
    <tr>
      <td>Grok 4</td>
      <td>-</td>
      <td>80.9</td>
      <td>70.3</td>
      <td>22.5</td>
      <td>40.7</td>
      <td>55.9</td>
      <td>64.8</td>
      <td>98.8</td>
      <td>93.9</td>
      <td>85.5</td>
      <td>87.5</td>
      <td>79.3</td>
    </tr>
  </tbody>
</table>

Note: Parts of the evaluation results are reproduced using the same settings.  
†: Evaluation results of Gemini 2.5 Flash (thinking) may be lower than real model performance, especially on MathVision, due to insufficient instruction following ability. 
## Deployment

> [!Note]
> Step3's API is accessible at https://platform.stepfun.com/, where we offer OpenAI-compatible API for you.

### Inference with Hugging Face Transformers

We introduce how to use our model at inference stage using transformers library. It is recommended to use python=3.10, torch>=2.1.0, and transformers=4.54.0 as the development environment.We currently only support bf16 inference, and multi-patch for image preprocessing is supported by default. This behavior is aligned with vllm and sglang.


```python
from transformers import AutoProcessor, AutoModelForCausalLM

key_mapping = {
    "^vision_model": "model.vision_model",
    r"^model(?!\.(language_model|vision_model))": "model.language_model",
    "vit_downsampler": "model.vit_downsampler",
    "vit_downsampler2": "model.vit_downsampler2",
    "vit_large_projector": "model.vit_large_projector",
}

model_path = "stepfun-ai/step3"

processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_path, 
                device_map="auto", torch_dtype="auto",trust_remote_code=True, 
                key_mapping=key_mapping)

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
            {"type": "text", "text": "What's in this picture?"}
        ]
    },
]

inputs = processor.apply_chat_template(
    messages, add_generation_prompt=True, tokenize=True,
    return_dict=True, return_tensors="pt"
).to(model.device)

generate_ids = model.generate(**inputs, max_new_tokens=32768, do_sample=False)
decoded = processor.decode(generate_ids[0, inputs["input_ids"].shape[-1] :], skip_special_tokens=True)

print(decoded)

```


### Inference with vLLM and SGLang


Our model checkpoints are stored in bf16 and block-fp8 format, you can find it on [Huggingface](https://huggingface.co/collections/stepfun-ai/step3-688a3d652dbb45d868f9d42d).

Currently, it is recommended to run Step3 on the following inference engines:

* vLLM
* SGLang

Deployment and Request examples for vLLM and SGLang can be found in the [Model Deployment Guide](docs/deploy_guidance.md).

## Contact Us
If you have any questions, please reach out at [contact@stepfun.com](mailto:contact@stepfun.com) .

## License
Both the code repository and the model weights are released under the [Apache License (Version 2.0)](./LICENSE).

## Citation
```
@misc{step3system,
      title={Step-3 is Large yet Affordable: Model-system Co-design for Cost-effective Decoding}, 
      author={StepFun Team},
      year={2025},
      eprint={2507.19427},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2507.19427}, 
}

@misc{step3blog,
      title={Step3: Cost-Effective Multimodal Intelligence}, 
      author={StepFun Team},
      url={https://stepfun.ai/research/step3}, 
}
```