Hugging Face's logo

AtomicChat
/
Kimi-K3

Image-Text-to-Text
Transformers
Safetensors
kimi_k3
feature-extraction
compressed-tensors
conversational
custom_code
8-bit precision
Model card Files
xet
 Community
worthant commited on
Commit
b8f1651
Β·
verified Β·
1 Parent(s): 73d6de1

mirror: card from moonshotai/Kimi-K3 + Atomic Chat notice

Browse files
Files changed (1) hide show
  1. README.md +686 -0
README.md ADDED
@@ -0,0 +1,686 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - compressed-tensors
4
+ - conversational
5
+ license: other
6
+ license_name: "kimi-k3"
7
+ library_name: transformers
8
+ pipeline_tag: image-text-to-text
9
+ ---
10
+
11
+ > ### Mirror
12
+ > This repository is a verbatim mirror of [`moonshotai/Kimi-K3`](https://huggingface.co/moonshotai/Kimi-K3), re-hosted by [Atomic Chat](https://atomic.chat) under the terms of the **Kimi K3 License**.
13
+ > All weights, configuration and the model card below are the work of **Moonshot AI**; we claim no ownership and add nothing. Original and full credit: https://huggingface.co/moonshotai/Kimi-K3
14
+
15
+
16
+ <div align="center">
17
+ <picture>
18
+ <img src="assets/kimi-logo.png" width="30%" alt="Kimi K3">
19
+ </picture>
20
+ </div>
21
+ <hr>
22
+ <div align="center" style="line-height:1">
23
+ <a href="https://www.kimi.com" target="_blank"><img alt="Chat" src="https://img.shields.io/badge/πŸ€–%20Chat-Kimi%20K3-ff6b6b?color=1783ff&logoColor=white"/></a>
24
+ <a href="https://www.moonshot.ai" target="_blank"><img alt="Homepage" src="https://img.shields.io/badge/Homepage-Moonshot%20AI-white?logo=Kimi&logoColor=white"/></a>
25
+ </div>
26
+
27
+ <div align="center" style="line-height: 1;">
28
+ <a href="https://huggingface.co/moonshotai" target="_blank"><img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Moonshot%20AI-ffc107?color=ffc107&logoColor=white"/></a>
29
+ <a href="https://twitter.com/kimi_moonshot" target="_blank"><img alt="Twitter Follow" src="https://img.shields.io/badge/Twitter-Kimi.ai-white?logo=x&logoColor=white"/></a>
30
+ <a href="https://discord.gg/TYU2fdJykW" target="_blank"><img alt="Discord" src="https://img.shields.io/badge/Discord-Kimi.ai-white?logo=discord&logoColor=white"/></a>
31
+ <a href="https://modelscope.cn/organization/moonshotai" target="_blank"><img alt="ModelScope" src="https://img.shields.io/badge/ModelScope-Moonshot%20AI-white?labelColor=rgb(99%2C%2074%2C%255)"/></a>
32
+ </div>
33
+ <div align="center" style="line-height: 1;">
34
+ <a href="https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE"><img alt="License" src="https://img.shields.io/badge/License-Kimi_K3-f5de53?&color=f5de53"/></a>
35
+ </div>
36
+
37
+
38
+ <p align="center">
39
+ πŸ“°&nbsp;&nbsp;<a href="https://www.kimi.com/blog/kimi-k3">Tech Blog</a> | &nbsp;&nbsp;&nbsp; <b>πŸ“„&nbsp;&nbsp;<a href="https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf">Full Report</a></b>
40
+ </p>
41
+
42
+
43
+
44
+ ## 1. Model Introduction
45
+
46
+ Kimi K3 is an open-weight, native multimodal agentic model and our most capable model to date. It is a 2.8T-parameter model built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), with native vision capabilities and a 1-million-token context window. It is the world's first open 3T-class model, designed for frontier intelligence across long-horizon coding, knowledge work, and reasoning.
47
+
48
+ ### Key Features
49
+ - **New Architecture**: Kimi K3 is built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), and scales up MoE sparsity with a Stable LatentMoE framework that activates 16 out of 896 experts β€” yielding an approximate 2.5Γ— improvement in overall scaling efficiency over Kimi K2.
50
+ - **Long-Horizon Coding**: Operating with minimal human oversight, Kimi K3 sustains long engineering sessions, navigates massive repositories, and orchestrates terminal tools β€” from GPU kernel optimization and compiler development to vision-in-the-loop game dev, CAD, and even chip design.
51
+ - **Agentic Knowledge Work**: Kimi K3 advances end-to-end knowledge work, producing deep research with interactive visualizations, widgets and dashboards, and motion design and video editing, powered by its native multimodal architecture.
52
+ - **Native Multimodality & Long Context**: Kimi K3 understands text, images, and video within the same model, and supports a 1-million-token context window.
53
+ - **Open Frontier Weights**: We release the full Kimi K3 model weights under the Kimi K3 License, making frontier intelligence openly available for research, deployment, and further innovation.
54
+ ## 2. Model Summary
55
+
56
+ <div align="center">
57
+ <table>
58
+ <tbody>
59
+ <tr>
60
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Architecture</strong></td>
61
+ <td align="center" style="vertical-align: middle; text-align: center">Mixture-of-Experts (MoE)</td>
62
+ </tr>
63
+ <tr>
64
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Total Parameters</strong></td>
65
+ <td align="center" style="vertical-align: middle; text-align: center">2.8T</td>
66
+ </tr>
67
+ <tr>
68
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Activated Parameters</strong></td>
69
+ <td align="center" style="vertical-align: middle; text-align: center">104B</td>
70
+ </tr>
71
+ <tr>
72
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Number of Layers</strong></td>
73
+ <td align="center" style="vertical-align: middle; text-align: center">93</td>
74
+ </tr>
75
+ <tr>
76
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Number of Dense Layers</strong></td>
77
+ <td align="center" style="vertical-align: middle; text-align: center">1</td>
78
+ </tr>
79
+ <tr>
80
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Attention-Layer Composition</strong></td>
81
+ <td align="center" style="vertical-align: middle; text-align: center">69 KDA + 24 Gated MLA</td>
82
+ </tr>
83
+ <tr>
84
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Attention Hidden Dimension</strong></td>
85
+ <td align="center" style="vertical-align: middle; text-align: center">7168</td>
86
+ </tr>
87
+ <tr>
88
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Number of Attention Heads</strong></td>
89
+ <td align="center" style="vertical-align: middle; text-align: center">96</td>
90
+ </tr>
91
+ <tr>
92
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Latent MoE Dimension</strong></td>
93
+ <td align="center" style="vertical-align: middle; text-align: center">3584</td>
94
+ </tr>
95
+ <tr>
96
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>MoE Hidden Dimension</strong> (per Expert)</td>
97
+ <td align="center" style="vertical-align: middle; text-align: center">3072</td>
98
+ </tr>
99
+ <tr>
100
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Number of Experts</strong></td>
101
+ <td align="center" style="vertical-align: middle; text-align: center">896</td>
102
+ </tr>
103
+ <tr>
104
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Selected Experts per Token</strong></td>
105
+ <td align="center" style="vertical-align: middle; text-align: center">16</td>
106
+ </tr>
107
+ <tr>
108
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Number of Shared Experts</strong></td>
109
+ <td align="center" style="vertical-align: middle; text-align: center">2</td>
110
+ </tr>
111
+ <tr>
112
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Vocabulary Size</strong></td>
113
+ <td align="center" style="vertical-align: middle; text-align: center">160K</td>
114
+ </tr>
115
+ <tr>
116
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Context Length</strong></td>
117
+ <td align="center" style="vertical-align: middle; text-align: center">1048576</td>
118
+ </tr>
119
+ <tr>
120
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Attention Mechanism</strong></td>
121
+ <td align="center" style="vertical-align: middle; text-align: center">KDA &amp; Gated MLA</td>
122
+ </tr>
123
+ <tr>
124
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Activation Function</strong></td>
125
+ <td align="center" style="vertical-align: middle; text-align: center">SiTU-GLU</td>
126
+ </tr>
127
+ <tr>
128
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Vision Encoder</strong></td>
129
+ <td align="center" style="vertical-align: middle; text-align: center">MoonViT-V2</td>
130
+ </tr>
131
+ <tr>
132
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Parameters of Vision Encoder</strong></td>
133
+ <td align="center" style="vertical-align: middle; text-align: center">401M</td>
134
+ </tr>
135
+ <tr>
136
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Quantization</strong></td>
137
+ <td align="center" style="vertical-align: middle; text-align: center">MXFP4 weights / MXFP8 activations<br>(quantization-aware training)</td>
138
+ </tr>
139
+ <tr>
140
+ <td align="center" style="vertical-align: middle; text-align: center"><strong>Modality</strong></td>
141
+ <td align="center" style="vertical-align: middle; text-align: center">Text, Image</td>
142
+ </tr>
143
+ </tbody>
144
+ </table>
145
+ </div>
146
+
147
+
148
+ ## 3. Evaluation Results
149
+
150
+ <div align="center">
151
+ <table>
152
+ <thead>
153
+ <tr>
154
+ <th align="center" style="text-align: center">Benchmark</th>
155
+ <th align="center" style="text-align: center"><sup>Kimi K3<br><sup>(max)</sup></sup></th>
156
+ <th align="center" style="text-align: center"><sup>Claude Fable 5<br><sup>(max, w/ fallback)</sup></sup></th>
157
+ <th align="center" style="text-align: center"><sup>GPT-5.6 Sol<br><sup>(max)</sup></sup></th>
158
+ <th align="center" style="text-align: center"><sup>Claude Opus 4.8<br><sup>(max)</sup></sup></th>
159
+ <th align="center" style="text-align: center"><sup>GPT-5.5<br><sup>(xhigh)</sup></sup></th>
160
+ <th align="center" style="text-align: center"><sup>GLM-5.2<br><sup>(max)</sup></sup></th>
161
+ </tr>
162
+ </thead>
163
+ <tbody>
164
+ <tr>
165
+ <td align="center" colspan=7 style="text-align: center"><strong>Reasoning &amp; Knowledge</strong></td>
166
+ </tr>
167
+ <tr>
168
+ <td align="center" style="vertical-align: middle; text-align: center">GPQA Diamond</td>
169
+ <td align="center" style="vertical-align: middle; text-align: center">93.5</td>
170
+ <td align="center" style="vertical-align: middle; text-align: center">92.6</td>
171
+ <td align="center" style="vertical-align: middle; text-align: center">94.1</td>
172
+ <td align="center" style="vertical-align: middle; text-align: center">91.0</td>
173
+ <td align="center" style="vertical-align: middle; text-align: center">93.5</td>
174
+ <td align="center" style="vertical-align: middle; text-align: center">91.2</td>
175
+ </tr>
176
+ <tr>
177
+ <td align="center" style="vertical-align: middle; text-align: center">CritPt</td>
178
+ <td align="center" style="vertical-align: middle; text-align: center">23.4</td>
179
+ <td align="center" style="vertical-align: middle; text-align: center">28.6</td>
180
+ <td align="center" style="vertical-align: middle; text-align: center">32.3</td>
181
+ <td align="center" style="vertical-align: middle; text-align: center">20.9</td>
182
+ <td align="center" style="vertical-align: middle; text-align: center">27.1</td>
183
+ <td align="center" style="vertical-align: middle; text-align: center">20.9</td>
184
+ </tr>
185
+ <tr>
186
+ <td align="center" style="vertical-align: middle; text-align: center">AA-LCR</td>
187
+ <td align="center" style="vertical-align: middle; text-align: center">74.7</td>
188
+ <td align="center" style="vertical-align: middle; text-align: center">70.0</td>
189
+ <td align="center" style="vertical-align: middle; text-align: center">73.7</td>
190
+ <td align="center" style="vertical-align: middle; text-align: center">67.7</td>
191
+ <td align="center" style="vertical-align: middle; text-align: center">74.3</td>
192
+ <td align="center" style="vertical-align: middle; text-align: center">71.3</td>
193
+ </tr>
194
+ <tr>
195
+ <td align="center" style="vertical-align: middle; text-align: center">HLE-Full</td>
196
+ <td align="center" style="vertical-align: middle; text-align: center">43.5 / 56.0</td>
197
+ <td align="center" style="vertical-align: middle; text-align: center">53.3 / 63.0</td>
198
+ <td align="center" style="vertical-align: middle; text-align: center">44.5 / 58.0</td>
199
+ <td align="center" style="vertical-align: middle; text-align: center">49.8 / 57.9</td>
200
+ <td align="center" style="vertical-align: middle; text-align: center">41.4 / 52.2</td>
201
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
202
+ </tr>
203
+ <tr>
204
+ <td align="center" colspan=7 style="text-align: center"><strong>Coding</strong></td>
205
+ </tr>
206
+ <tr>
207
+ <td align="center" style="vertical-align: middle; text-align: center">DeepSWE</td>
208
+ <td align="center" style="vertical-align: middle; text-align: center">67.5</td>
209
+ <td align="center" style="vertical-align: middle; text-align: center">70.0</td>
210
+ <td align="center" style="vertical-align: middle; text-align: center">73.0</td>
211
+ <td align="center" style="vertical-align: middle; text-align: center">59.0</td>
212
+ <td align="center" style="vertical-align: middle; text-align: center">67.0</td>
213
+ <td align="center" style="vertical-align: middle; text-align: center">46.2</td>
214
+ </tr>
215
+ <tr>
216
+ <td align="center" style="vertical-align: middle; text-align: center">ProgramBench</td>
217
+ <td align="center" style="vertical-align: middle; text-align: center">77.8</td>
218
+ <td align="center" style="vertical-align: middle; text-align: center">76.8</td>
219
+ <td align="center" style="vertical-align: middle; text-align: center">77.6</td>
220
+ <td align="center" style="vertical-align: middle; text-align: center">71.9</td>
221
+ <td align="center" style="vertical-align: middle; text-align: center">70.8</td>
222
+ <td align="center" style="vertical-align: middle; text-align: center">63.7</td>
223
+ </tr>
224
+ <tr>
225
+ <td align="center" style="vertical-align: middle; text-align: center">Terminal-Bench 2.1</td>
226
+ <td align="center" style="vertical-align: middle; text-align: center">88.3</td>
227
+ <td align="center" style="vertical-align: middle; text-align: center">88.0</td>
228
+ <td align="center" style="vertical-align: middle; text-align: center">88.8</td>
229
+ <td align="center" style="vertical-align: middle; text-align: center">84.6</td>
230
+ <td align="center" style="vertical-align: middle; text-align: center">83.4</td>
231
+ <td align="center" style="vertical-align: middle; text-align: center">82.7</td>
232
+ </tr>
233
+ <tr>
234
+ <td align="center" style="vertical-align: middle; text-align: center">FrontierSWE</td>
235
+ <td align="center" style="vertical-align: middle; text-align: center">81.2</td>
236
+ <td align="center" style="vertical-align: middle; text-align: center">86.6</td>
237
+ <td align="center" style="vertical-align: middle; text-align: center">71.3</td>
238
+ <td align="center" style="vertical-align: middle; text-align: center">66.7</td>
239
+ <td align="center" style="vertical-align: middle; text-align: center">64.9</td>
240
+ <td align="center" style="vertical-align: middle; text-align: center">67.3</td>
241
+ </tr>
242
+ <tr>
243
+ <td align="center" style="vertical-align: middle; text-align: center">SWE-Marathon</td>
244
+ <td align="center" style="vertical-align: middle; text-align: center">42.0</td>
245
+ <td align="center" style="vertical-align: middle; text-align: center">35.0</td>
246
+ <td align="center" style="vertical-align: middle; text-align: center">39.0</td>
247
+ <td align="center" style="vertical-align: middle; text-align: center">40.0</td>
248
+ <td align="center" style="vertical-align: middle; text-align: center">14.0</td>
249
+ <td align="center" style="vertical-align: middle; text-align: center">13.0</td>
250
+ </tr>
251
+ <tr>
252
+ <td align="center" style="vertical-align: middle; text-align: center">PostTrainBench</td>
253
+ <td align="center" style="vertical-align: middle; text-align: center">36.6</td>
254
+ <td align="center" style="vertical-align: middle; text-align: center">41.4</td>
255
+ <td align="center" style="vertical-align: middle; text-align: center">34.6</td>
256
+ <td align="center" style="vertical-align: middle; text-align: center">34.1</td>
257
+ <td align="center" style="vertical-align: middle; text-align: center">28.4</td>
258
+ <td align="center" style="vertical-align: middle; text-align: center">34.3</td>
259
+ </tr>
260
+ <tr>
261
+ <td align="center" style="vertical-align: middle; text-align: center">MLS-Bench-Lite</td>
262
+ <td align="center" style="vertical-align: middle; text-align: center">48.3</td>
263
+ <td align="center" style="vertical-align: middle; text-align: center">49.9</td>
264
+ <td align="center" style="vertical-align: middle; text-align: center">46.2</td>
265
+ <td align="center" style="vertical-align: middle; text-align: center">42.8</td>
266
+ <td align="center" style="vertical-align: middle; text-align: center">35.5</td>
267
+ <td align="center" style="vertical-align: middle; text-align: center">40.4</td>
268
+ </tr>
269
+ <tr>
270
+ <td align="center" style="vertical-align: middle; text-align: center">SciCode</td>
271
+ <td align="center" style="vertical-align: middle; text-align: center">58.7</td>
272
+ <td align="center" style="vertical-align: middle; text-align: center">60.2</td>
273
+ <td align="center" style="vertical-align: middle; text-align: center">56.1</td>
274
+ <td align="center" style="vertical-align: middle; text-align: center">53.5</td>
275
+ <td align="center" style="vertical-align: middle; text-align: center">56.1</td>
276
+ <td align="center" style="vertical-align: middle; text-align: center">50.5</td>
277
+ </tr>
278
+ <tr>
279
+ <td align="center" style="vertical-align: middle; text-align: center">Kimi Code Bench 2.0</td>
280
+ <td align="center" style="vertical-align: middle; text-align: center">72.9</td>
281
+ <td align="center" style="vertical-align: middle; text-align: center">76.9</td>
282
+ <td align="center" style="vertical-align: middle; text-align: center">64.8</td>
283
+ <td align="center" style="vertical-align: middle; text-align: center">71.7</td>
284
+ <td align="center" style="vertical-align: middle; text-align: center">69.0</td>
285
+ <td align="center" style="vertical-align: middle; text-align: center">64.2</td>
286
+ </tr>
287
+ <tr>
288
+ <td align="center" colspan=7 style="text-align: center"><strong>Agentic</strong></td>
289
+ </tr>
290
+ <tr>
291
+ <td align="center" style="vertical-align: middle; text-align: center">BrowseComp</td>
292
+ <td align="center" style="vertical-align: middle; text-align: center">91.2</td>
293
+ <td align="center" style="vertical-align: middle; text-align: center">88.0</td>
294
+ <td align="center" style="vertical-align: middle; text-align: center">90.4</td>
295
+ <td align="center" style="vertical-align: middle; text-align: center">84.3</td>
296
+ <td align="center" style="vertical-align: middle; text-align: center">84.4</td>
297
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
298
+ </tr>
299
+ <tr>
300
+ <td align="center" style="vertical-align: middle; text-align: center">DeepSearchQA (F1)</td>
301
+ <td align="center" style="vertical-align: middle; text-align: center">95.0</td>
302
+ <td align="center" style="vertical-align: middle; text-align: center">94.2</td>
303
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
304
+ <td align="center" style="vertical-align: middle; text-align: center">93.1</td>
305
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
306
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
307
+ </tr>
308
+ <tr>
309
+ <td align="center" style="vertical-align: middle; text-align: center">ResearchRubrics</td>
310
+ <td align="center" style="vertical-align: middle; text-align: center">76.2</td>
311
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
312
+ <td align="center" style="vertical-align: middle; text-align: center">73.8</td>
313
+ <td align="center" style="vertical-align: middle; text-align: center">73.5</td>
314
+ <td align="center" style="vertical-align: middle; text-align: center">64.0</td>
315
+ <td align="center" style="vertical-align: middle; text-align: center">71.1</td>
316
+ </tr>
317
+ <tr>
318
+ <td align="center" style="vertical-align: middle; text-align: center">GDPval-AA v2 (Elo)</td>
319
+ <td align="center" style="vertical-align: middle; text-align: center">1686</td>
320
+ <td align="center" style="vertical-align: middle; text-align: center">1747</td>
321
+ <td align="center" style="vertical-align: middle; text-align: center">1736</td>
322
+ <td align="center" style="vertical-align: middle; text-align: center">1593</td>
323
+ <td align="center" style="vertical-align: middle; text-align: center">1491</td>
324
+ <td align="center" style="vertical-align: middle; text-align: center">1510</td>
325
+ </tr>
326
+ <tr>
327
+ <td align="center" style="vertical-align: middle; text-align: center">Toolathlon-Verified</td>
328
+ <td align="center" style="vertical-align: middle; text-align: center">76.5</td>
329
+ <td align="center" style="vertical-align: middle; text-align: center">77.9</td>
330
+ <td align="center" style="vertical-align: middle; text-align: center">74.9</td>
331
+ <td align="center" style="vertical-align: middle; text-align: center">76.2</td>
332
+ <td align="center" style="vertical-align: middle; text-align: center">73.5</td>
333
+ <td align="center" style="vertical-align: middle; text-align: center">59.9</td>
334
+ </tr>
335
+ <tr>
336
+ <td align="center" style="vertical-align: middle; text-align: center">MCPMark-Verified</td>
337
+ <td align="center" style="vertical-align: middle; text-align: center">94.5</td>
338
+ <td align="center" style="vertical-align: middle; text-align: center">87.4</td>
339
+ <td align="center" style="vertical-align: middle; text-align: center">92.9</td>
340
+ <td align="center" style="vertical-align: middle; text-align: center">76.4</td>
341
+ <td align="center" style="vertical-align: middle; text-align: center">92.9</td>
342
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
343
+ </tr>
344
+ <tr>
345
+ <td align="center" style="vertical-align: middle; text-align: center">MCP-Atlas</td>
346
+ <td align="center" style="vertical-align: middle; text-align: center">84.2</td>
347
+ <td align="center" style="vertical-align: middle; text-align: center">84.7</td>
348
+ <td align="center" style="vertical-align: middle; text-align: center">83.6</td>
349
+ <td align="center" style="vertical-align: middle; text-align: center">83.6</td>
350
+ <td align="center" style="vertical-align: middle; text-align: center">82.8</td>
351
+ <td align="center" style="vertical-align: middle; text-align: center">82.6</td>
352
+ </tr>
353
+ <tr>
354
+ <td align="center" style="vertical-align: middle; text-align: center">AutomationBench</td>
355
+ <td align="center" style="vertical-align: middle; text-align: center">30.8</td>
356
+ <td align="center" style="vertical-align: middle; text-align: center">29.1</td>
357
+ <td align="center" style="vertical-align: middle; text-align: center">29.7</td>
358
+ <td align="center" style="vertical-align: middle; text-align: center">27.2</td>
359
+ <td align="center" style="vertical-align: middle; text-align: center">22.7</td>
360
+ <td align="center" style="vertical-align: middle; text-align: center">12.9</td>
361
+ </tr>
362
+ <tr>
363
+ <td align="center" style="vertical-align: middle; text-align: center">JobBench</td>
364
+ <td align="center" style="vertical-align: middle; text-align: center">54.3</td>
365
+ <td align="center" style="vertical-align: middle; text-align: center">57.4</td>
366
+ <td align="center" style="vertical-align: middle; text-align: center">45.4</td>
367
+ <td align="center" style="vertical-align: middle; text-align: center">48.4</td>
368
+ <td align="center" style="vertical-align: middle; text-align: center">38.3</td>
369
+ <td align="center" style="vertical-align: middle; text-align: center">43.4</td>
370
+ </tr>
371
+ <tr>
372
+ <td align="center" style="vertical-align: middle; text-align: center">AA-Briefcase (Elo)</td>
373
+ <td align="center" style="vertical-align: middle; text-align: center">1548</td>
374
+ <td align="center" style="vertical-align: middle; text-align: center">1583</td>
375
+ <td align="center" style="vertical-align: middle; text-align: center">1495</td>
376
+ <td align="center" style="vertical-align: middle; text-align: center">1354</td>
377
+ <td align="center" style="vertical-align: middle; text-align: center">1158</td>
378
+ <td align="center" style="vertical-align: middle; text-align: center">1260</td>
379
+ </tr>
380
+ <tr>
381
+ <td align="center" style="vertical-align: middle; text-align: center">Agents' Last Exam</td>
382
+ <td align="center" style="vertical-align: middle; text-align: center">28.3</td>
383
+ <td align="center" style="vertical-align: middle; text-align: center">25.7<sup>†</sup></td>
384
+ <td align="center" style="vertical-align: middle; text-align: center">29.6</td>
385
+ <td align="center" style="vertical-align: middle; text-align: center">27.0</td>
386
+ <td align="center" style="vertical-align: middle; text-align: center">26.6</td>
387
+ <td align="center" style="vertical-align: middle; text-align: center">20.4</td>
388
+ </tr>
389
+ <tr>
390
+ <td align="center" style="vertical-align: middle; text-align: center">APEX-Agents</td>
391
+ <td align="center" style="vertical-align: middle; text-align: center">41.0</td>
392
+ <td align="center" style="vertical-align: middle; text-align: center">43.3</td>
393
+ <td align="center" style="vertical-align: middle; text-align: center">39.9</td>
394
+ <td align="center" style="vertical-align: middle; text-align: center">39.4</td>
395
+ <td align="center" style="vertical-align: middle; text-align: center">38.5</td>
396
+ <td align="center" style="vertical-align: middle; text-align: center">35.6</td>
397
+ </tr>
398
+ <tr>
399
+ <td align="center" style="vertical-align: middle; text-align: center">OfficeQA Pro</td>
400
+ <td align="center" style="vertical-align: middle; text-align: center">63.3</td>
401
+ <td align="center" style="vertical-align: middle; text-align: center">69.9</td>
402
+ <td align="center" style="vertical-align: middle; text-align: center">63.2</td>
403
+ <td align="center" style="vertical-align: middle; text-align: center">63.9</td>
404
+ <td align="center" style="vertical-align: middle; text-align: center">60.9</td>
405
+ <td align="center" style="vertical-align: middle; text-align: center">41.4</td>
406
+ </tr>
407
+ <tr>
408
+ <td align="center" style="vertical-align: middle; text-align: center">SpreadsheetBench 2</td>
409
+ <td align="center" style="vertical-align: middle; text-align: center">34.8</td>
410
+ <td align="center" style="vertical-align: middle; text-align: center">34.7</td>
411
+ <td align="center" style="vertical-align: middle; text-align: center">32.4</td>
412
+ <td align="center" style="vertical-align: middle; text-align: center">31.6</td>
413
+ <td align="center" style="vertical-align: middle; text-align: center">29.1</td>
414
+ <td align="center" style="vertical-align: middle; text-align: center">28.1</td>
415
+ </tr>
416
+ <tr>
417
+ <td align="center" style="vertical-align: middle; text-align: center">OSWorld-Verified</td>
418
+ <td align="center" style="vertical-align: middle; text-align: center">84.8</td>
419
+ <td align="center" style="vertical-align: middle; text-align: center">85.0</td>
420
+ <td align="center" style="vertical-align: middle; text-align: center">83.0</td>
421
+ <td align="center" style="vertical-align: middle; text-align: center">83.4</td>
422
+ <td align="center" style="vertical-align: middle; text-align: center">79.0</td>
423
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
424
+ </tr>
425
+ <tr>
426
+ <td align="center" style="vertical-align: middle; text-align: center">OSWorld 2.0</td>
427
+ <td align="center" style="vertical-align: middle; text-align: center">58.3</td>
428
+ <td align="center" style="vertical-align: middle; text-align: center">66.1</td>
429
+ <td align="center" style="vertical-align: middle; text-align: center">62.6</td>
430
+ <td align="center" style="vertical-align: middle; text-align: center">55.7</td>
431
+ <td align="center" style="vertical-align: middle; text-align: center">49.5</td>
432
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
433
+ </tr>
434
+ <tr>
435
+ <td align="center" style="vertical-align: middle; text-align: center">SaaS-Bench</td>
436
+ <td align="center" style="vertical-align: middle; text-align: center">60.1</td>
437
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
438
+ <td align="center" style="vertical-align: middle; text-align: center">61.4</td>
439
+ <td align="center" style="vertical-align: middle; text-align: center">56.1</td>
440
+ <td align="center" style="vertical-align: middle; text-align: center">43.8</td>
441
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
442
+ </tr>
443
+ <tr>
444
+ <td align="center" style="vertical-align: middle; text-align: center">τ³-Banking</td>
445
+ <td align="center" style="vertical-align: middle; text-align: center">33.4</td>
446
+ <td align="center" style="vertical-align: middle; text-align: center">26.8</td>
447
+ <td align="center" style="vertical-align: middle; text-align: center">33.0</td>
448
+ <td align="center" style="vertical-align: middle; text-align: center">27.6</td>
449
+ <td align="center" style="vertical-align: middle; text-align: center">31.3</td>
450
+ <td align="center" style="vertical-align: middle; text-align: center">26.8</td>
451
+ </tr>
452
+ <tr>
453
+ <td align="center" style="vertical-align: middle; text-align: center">Harvey Lab-AA</td>
454
+ <td align="center" style="vertical-align: middle; text-align: center">94.6</td>
455
+ <td align="center" style="vertical-align: middle; text-align: center">93.6</td>
456
+ <td align="center" style="vertical-align: middle; text-align: center">87.2</td>
457
+ <td align="center" style="vertical-align: middle; text-align: center">91.1</td>
458
+ <td align="center" style="vertical-align: middle; text-align: center">86.3</td>
459
+ <td align="center" style="vertical-align: middle; text-align: center">91.0</td>
460
+ </tr>
461
+ <tr>
462
+ <td align="center" style="vertical-align: middle; text-align: center">CorpFin v2</td>
463
+ <td align="center" style="vertical-align: middle; text-align: center">71.6</td>
464
+ <td align="center" style="vertical-align: middle; text-align: center">71.8</td>
465
+ <td align="center" style="vertical-align: middle; text-align: center">64.4</td>
466
+ <td align="center" style="vertical-align: middle; text-align: center">66.7</td>
467
+ <td align="center" style="vertical-align: middle; text-align: center">68.4</td>
468
+ <td align="center" style="vertical-align: middle; text-align: center">66.1</td>
469
+ </tr>
470
+ <tr>
471
+ <td align="center" style="vertical-align: middle; text-align: center">Finance Agent v2</td>
472
+ <td align="center" style="vertical-align: middle; text-align: center">54.4</td>
473
+ <td align="center" style="vertical-align: middle; text-align: center">56.3</td>
474
+ <td align="center" style="vertical-align: middle; text-align: center">53.8</td>
475
+ <td align="center" style="vertical-align: middle; text-align: center">53.9</td>
476
+ <td align="center" style="vertical-align: middle; text-align: center">51.8</td>
477
+ <td align="center" style="vertical-align: middle; text-align: center">49.7</td>
478
+ </tr>
479
+ <tr>
480
+ <td align="center" style="vertical-align: middle; text-align: center">Legal Research Bench</td>
481
+ <td align="center" style="vertical-align: middle; text-align: center">44.2</td>
482
+ <td align="center" style="vertical-align: middle; text-align: center">49.5</td>
483
+ <td align="center" style="vertical-align: middle; text-align: center">48.1</td>
484
+ <td align="center" style="vertical-align: middle; text-align: center">43.8</td>
485
+ <td align="center" style="vertical-align: middle; text-align: center">40.4</td>
486
+ <td align="center" style="vertical-align: middle; text-align: center">31.3</td>
487
+ </tr>
488
+ <tr>
489
+ <td align="center" colspan=7 style="text-align: center"><strong>Vision</strong></td>
490
+ </tr>
491
+ <tr>
492
+ <td align="center" style="vertical-align: middle; text-align: center">WorldVQA ForceAnswer</td>
493
+ <td align="center" style="vertical-align: middle; text-align: center">51.0</td>
494
+ <td align="center" style="vertical-align: middle; text-align: center">56.7</td>
495
+ <td align="center" style="vertical-align: middle; text-align: center">41.8</td>
496
+ <td align="center" style="vertical-align: middle; text-align: center">39.1</td>
497
+ <td align="center" style="vertical-align: middle; text-align: center">38.5</td>
498
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
499
+ </tr>
500
+ <tr>
501
+ <td align="center" style="vertical-align: middle; text-align: center">OmniDocBench</td>
502
+ <td align="center" style="vertical-align: middle; text-align: center">91.1</td>
503
+ <td align="center" style="vertical-align: middle; text-align: center">89.8</td>
504
+ <td align="center" style="vertical-align: middle; text-align: center">85.8</td>
505
+ <td align="center" style="vertical-align: middle; text-align: center">87.9</td>
506
+ <td align="center" style="vertical-align: middle; text-align: center">89.4</td>
507
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
508
+ </tr>
509
+ <tr>
510
+ <td align="center" style="vertical-align: middle; text-align: center">PerceptionBench</td>
511
+ <td align="center" style="vertical-align: middle; text-align: center">58.5</td>
512
+ <td align="center" style="vertical-align: middle; text-align: center">57.2</td>
513
+ <td align="center" style="vertical-align: middle; text-align: center">59.7</td>
514
+ <td align="center" style="vertical-align: middle; text-align: center">47.2</td>
515
+ <td align="center" style="vertical-align: middle; text-align: center">55.8</td>
516
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
517
+ </tr>
518
+ <tr>
519
+ <td align="center" style="vertical-align: middle; text-align: center">Video-MME (w. sub)</td>
520
+ <td align="center" style="vertical-align: middle; text-align: center">90.0</td>
521
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
522
+ <td align="center" style="vertical-align: middle; text-align: center">89.5</td>
523
+ <td align="center" style="vertical-align: middle; text-align: center">86.0</td>
524
+ <td align="center" style="vertical-align: middle; text-align: center">89.3</td>
525
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
526
+ </tr>
527
+ <tr>
528
+ <td align="center" style="vertical-align: middle; text-align: center">MMVU</td>
529
+ <td align="center" style="vertical-align: middle; text-align: center">82.1</td>
530
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
531
+ <td align="center" style="vertical-align: middle; text-align: center">81.2</td>
532
+ <td align="center" style="vertical-align: middle; text-align: center">79.2</td>
533
+ <td align="center" style="vertical-align: middle; text-align: center">81.7</td>
534
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
535
+ </tr>
536
+ <tr>
537
+ <td align="center" style="vertical-align: middle; text-align: center">BabyVision w/ python</td>
538
+ <td align="center" style="vertical-align: middle; text-align: center">85.7</td>
539
+ <td align="center" style="vertical-align: middle; text-align: center">90.5</td>
540
+ <td align="center" style="vertical-align: middle; text-align: center">88.9</td>
541
+ <td align="center" style="vertical-align: middle; text-align: center">81.2</td>
542
+ <td align="center" style="vertical-align: middle; text-align: center">83.6</td>
543
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
544
+ </tr>
545
+ <tr>
546
+ <td align="center" style="vertical-align: middle; text-align: center">MMMU-Pro</td>
547
+ <td align="center" style="vertical-align: middle; text-align: center">81.6 / 83.4</td>
548
+ <td align="center" style="vertical-align: middle; text-align: center">81.2 / 86.5</td>
549
+ <td align="center" style="vertical-align: middle; text-align: center">83.0 / 84.6</td>
550
+ <td align="center" style="vertical-align: middle; text-align: center">78.9 / 82.7</td>
551
+ <td align="center" style="vertical-align: middle; text-align: center">81.2 / 83.2</td>
552
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
553
+ </tr>
554
+ <tr>
555
+ <td align="center" style="vertical-align: middle; text-align: center">CharXiv (RQ)</td>
556
+ <td align="center" style="vertical-align: middle; text-align: center">84.8 / 91.3</td>
557
+ <td align="center" style="vertical-align: middle; text-align: center">88.9 / 93.5</td>
558
+ <td align="center" style="vertical-align: middle; text-align: center">84.6 / 89.1</td>
559
+ <td align="center" style="vertical-align: middle; text-align: center">80.5 / 89.9</td>
560
+ <td align="center" style="vertical-align: middle; text-align: center">84.1 / 89.0</td>
561
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
562
+ </tr>
563
+ <tr>
564
+ <td align="center" style="vertical-align: middle; text-align: center">MathVision</td>
565
+ <td align="center" style="vertical-align: middle; text-align: center">94.3 / 97.8</td>
566
+ <td align="center" style="vertical-align: middle; text-align: center">94.8 / 98.6</td>
567
+ <td align="center" style="vertical-align: middle; text-align: center">95.8 / 97.8</td>
568
+ <td align="center" style="vertical-align: middle; text-align: center">86.7 / 97.1</td>
569
+ <td align="center" style="vertical-align: middle; text-align: center">92.2 / 96.8</td>
570
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
571
+ </tr>
572
+ <tr>
573
+ <td align="center" style="vertical-align: middle; text-align: center">ZeroBench (pass@5)</td>
574
+ <td align="center" style="vertical-align: middle; text-align: center">23.0 / 41.0</td>
575
+ <td align="center" style="vertical-align: middle; text-align: center">23.0 / 46.0</td>
576
+ <td align="center" style="vertical-align: middle; text-align: center">17.0 / 35.0</td>
577
+ <td align="center" style="vertical-align: middle; text-align: center">17.0 / 34.0</td>
578
+ <td align="center" style="vertical-align: middle; text-align: center">22.0 / 41.0</td>
579
+ <td align="center" style="vertical-align: middle; text-align: center">β€”</td>
580
+ </tr>
581
+ </tbody>
582
+ </table>
583
+ </div>
584
+
585
+ <details>
586
+ <summary><b>Footnotes</b></summary>
587
+
588
+ All Kimi K3 results are obtained with reasoning effort set to 'max' and temperature = 1.0. For single-step tasks, such as GPQA Diamond, HLE-Full, and vision benchmarks without tools, we set top-p = 0.95; for agentic tasks, we set top-p = 1.0. For HLE-Full, MMMU-Pro, CharXiv (RQ), MathVision, and ZeroBench, each cell reports the scores without and with tool augmentation (general tools for HLE-Full, Python for the vision benchmarks), in that order.
589
+
590
+ 1. **Reasoning & knowledge benchmarks**
591
+ - **CritPt and AA-LCR.** Scores are cited from [Artificial Analysis](https://artificialanalysis.ai/) as of July 23, 2026.
592
+ 2. **Coding benchmarks**
593
+ - **DeepSWE.** Kimi K3 is evaluated with the Kimi Code harness. The GLM-5.2 score is taken from the [GLM-5.2 release blog](https://z.ai/blog/glm-5.2); all remaining scores are from the official [DeepSWE leaderboard](https://deepswe.datacurve.ai/), under which Kimi K3 attains 67.3 with the mini-SWE-agent harness. We report the DeepSWE v1.1 tasks.
594
+ - **Terminal-Bench 2.1.** Kimi K3 is evaluated with the Kimi Code harness. For all other models, we report the best score across harnesses: GLM-5.2 with Claude Code ([GLM-5.2 release blog](https://z.ai/blog/glm-5.2)); Claude Opus 4.8 and Claude Fable 5 with Terminus 2 ([Artificial Analysis](https://artificialanalysis.ai/evaluations/terminalbench-v2-1)); GPT-5.5 and GPT-5.6 Sol with Codex ([OpenAI](https://openai.com/index/previewing-gpt-5-6-sol/)).
595
+ - **ProgramBench.** Kimi K3 is evaluated with the Kimi Code harness. The GLM-5.2 score is from the [GLM-5.2 release blog](https://z.ai/blog/glm-5.2); all other scores are from [Vals AI](https://www.vals.ai/benchmarks/programbench).
596
+ - **SWE-Marathon.** Kimi K3, Claude Opus 4.8, and Claude Fable 5 are evaluated with the Claude Code harness; GPT-5.6 Sol is evaluated with the Codex harness. The GLM-5.2 score is from the [GLM-5.2 release blog](https://z.ai/blog/glm-5.2). Our evaluation is based on an H20-calibrated branch of the [official tasks](https://www.swe-marathon.org/) as of July 9, 2026, prior to the final v1.1 release: the Docker images, performance gates, and reference oracles for the GPU tasks have been recalibrated for H20, while the correctness and anti-cheat validators remain unchanged. Additionally, Claude Fable 5 hit fallbacks on 35% of the tasks in our evaluation, which may have negatively impacted its measured performance.
597
+ - **FrontierSWE.** Kimi K3 is evaluated with the Kimi Code harness and GPT-5.6 Sol with the Codex harness; all other results are from [FrontierSWE](https://www.frontierswe.com/). Dominance scores are recomputed from the raw scores using the official evaluation script and are current as of July 16, 2026.
598
+ - **PostTrainBench.** Scores for GLM-5.2, GPT-5.5, and Claude Opus 4.8 are adopted from the official [PostTrainBench](https://posttrainbench.com/) results. Kimi K3, Claude Fable 5, and GPT-5.6 Sol are evaluated with the official Harbor implementation at maximum reasoning effort, averaged over three runs on H20 GPUs (instead of H100 in the official setting) β€” Kimi K3 and Claude Fable 5 with the Claude Code harness, and GPT-5.6 Sol with the Codex harness.
599
+ - **MLS-Bench-Lite.** Kimi K3 is evaluated with the Kimi Code harness; GLM-5.2 and the Claude models with the Claude Code harness; GPT-5.5 and GPT-5.6 Sol with the Codex harness.
600
+ - **SciCode.** Scores are cited from [Artificial Analysis](https://artificialanalysis.ai/) as of July 23, 2026.
601
+ - **Kimi Code Bench 2.0 (in-house).** Kimi K3 is evaluated with the Kimi Code harness (it attains 73.7 with the Claude Code harness); GLM-5.2, Claude Opus 4.8, and Claude Fable 5 with the Claude Code harness; GPT-5.5 and GPT-5.6 Sol with the Codex harness. All models are evaluated at maximum reasoning effort, except GPT-5.5, which uses the "xhigh" setting. As the benchmark includes cybersecurity and safety-related tasks, we also disclose the fraction of refused or fallback tasks: Claude Fable 5 hit 13 fallbacks and 1 refusal out of 80 tasks; 10 refusals out of 80 tasks entered GPT-5.6 Sol's cyber guard; GPT-5.5 had 3 refusals out of 80 tasks.
602
+ 3. **Agentic benchmarks**
603
+ - **OfficeQA Pro.** Each test case provides the agent with the entire PDF corpus, with all PDFs rendered as images and no machine-readable text available.
604
+ - **OfficeQA Pro and SpreadsheetBench 2.** Kimi K3, GLM-5.2, Claude Opus 4.8, and Claude Fable 5 are evaluated with the Claude Code harness; GPT-5.5 and GPT-5.6 Sol are evaluated with the Codex harness.
605
+ - **MCP-Atlas.** All models are evaluated on the 500-task public subset with a 100-turn limit, using Gemini 3.1 Pro as the judge.
606
+ - **AutomationBench.** All models are evaluated on the 600-task public subset, following the official GitHub setup in all other respects.
607
+ - **BrowseComp.** We adopt a context-compaction strategy triggered at 300K tokens. When evaluated with the full 1M-token context window and no context management, Kimi K3 achieves a score of 90.4. The results of Claude Fable 5, Claude Opus 4.8, GPT-5.6 Sol, and GPT-5.5 are cited from [Anthropic](https://www.anthropic.com/news/claude-fable-5-mythos-5) and [OpenAI](https://openai.com/index/gpt-5-6/).
608
+ - **GDPval-AA v2, AA-Briefcase, τ³-Banking, Harvey Lab-AA, and APEX-Agents.** Scores are cited from [Artificial Analysis](https://artificialanalysis.ai/) and the [APEX-Agents leaderboard](https://www.mercor.com/apex/apex-agents-leaderboard/) as of July 23, 2026. For Harvey Lab-AA, we report the criterion pass rate.
609
+ - **CorpFin v2, Finance Agent v2, and Legal Research Bench.** Scores are cited from [Vals AI](https://www.vals.ai/).
610
+ - **Agents' Last Exam.** Scores are cited from the [official leaderboard](https://agents-last-exam.org/leaderboard) as of July 23, 2026; we report the leaderboard's primary pass-rate metric. On the leaderboard, each model is paired with a specific harness: Kimi K3 with Kimi Code; GPT-5.6 Sol and GPT-5.5 with Codex; Claude Fable 5, Claude Opus 4.8, and GLM-5.2 with Claude Code. <sup>†</sup> The Claude Fable 5 entry runs at xhigh effort with 40% of tasks annotated as downgraded.
611
+ 4. **Multimodal benchmarks**
612
+ - Except for ZeroBench, which follows the official setting and is run five times, all multimodal scores are averaged over three runs. MMMU-Pro is evaluated following the official protocol, preserving the original input order and prepending images to the text input.
613
+ - **PerceptionBench** is an in-house benchmark that focuses on atomic visual perception capabilities.
614
+
615
+ </details>
616
+
617
+ ## 4. Native MXFP4 Quantization
618
+
619
+ Kimi K3 applies quantization-aware training from the SFT stage onward, using MXFP4 weights with MXFP8 activations for broad hardware compatibility.
620
+
621
+ ## 5. Deployment
622
+
623
+ > [!Note]
624
+ > You can access Kimi K3's API on https://platform.kimi.ai by selecting `kimi-k3`, and we provide OpenAI/Anthropic-compatible API for you. Currently, Kimi K3 is recommended to run on the following inference engines:
625
+
626
+ - [vLLM](https://github.com/vllm-project/vllm) β€” see [recipes](https://recipes.vllm.ai/moonshotai/Kimi-K3)
627
+ - [SGLang](https://github.com/sgl-project/sglang) β€” see [cookbook](https://docs.sglang.io/cookbook/autoregressive/Moonshotai/Kimi-K3)
628
+ - [TokenSpeed](https://github.com/lightseekorg/tokenspeed) β€” see [recipes](https://lightseek.org/tokenspeed/recipes/models#kimi-k3)
629
+
630
+ ---
631
+ ## 6. Model Usage
632
+
633
+ Kimi K3 always has thinking enabled, and will return `reasoning_content`. Thinking effort is configured with the top-level `reasoning_effort` request field, which supports `"low"`, `"high"`, and `"max"` (default `"max"`).
634
+
635
+ Kimi K3 was trained in the preserved thinking history mode. For multi-turn conversations and tool calls, Kimi K3 requires the complete assistant message returned by the API to be passed back to `messages` as-is β€” including `reasoning_content` and `tool_calls`, not just `content`:
636
+
637
+ ```python
638
+ import openai
639
+
640
+ def chat_with_preserved_thinking(client: openai.OpenAI, model_name: str):
641
+ messages = [
642
+ {
643
+ "role": "user",
644
+ "content": "Tell me three random numbers."
645
+ },
646
+ {
647
+ "role": "assistant",
648
+ "reasoning_content": "I'll start by listing five numbers: 473, 921, 235, 215, 222, and I'll tell you the first three.",
649
+ "content": "473, 921, 235"
650
+ },
651
+ {
652
+ "role": "user",
653
+ "content": "What are the other two numbers you have in mind?"
654
+ }
655
+ ]
656
+
657
+ response = client.chat.completions.create(
658
+ model=model_name,
659
+ messages=messages,
660
+ stream=False,
661
+ max_tokens=4096,
662
+ reasoning_effort="max",
663
+ )
664
+ # the assistant should mention 215 and 222 that appear in the prior reasoning content
665
+ print(f"response: {response.choices[0].message.reasoning}")
666
+ return response.choices[0].message.content
667
+ ```
668
+
669
+ For full guides and examples (vision input, structured output, partial mode, tool choice, dynamic tool loading, context caching), see the [Kimi K3 Quickstart](https://platform.kimi.ai/docs/guide/kimi-k3-quickstart) and [Thinking Effort](https://platform.kimi.ai/docs/guide/use-thinking-effort).
670
+
671
+ ### Coding Agent Framework
672
+
673
+ Kimi K3 works best with [Kimi Code CLI](https://www.kimi.com/code) as its agent framework. We warmly invite you to give it a try β€” run Kimi Code in your terminal and select Kimi K3 using the `/model` command. We hope you enjoy building with Kimi K3, and we would love to hear your feedback!
674
+
675
+
676
+ ---
677
+
678
+ ## 7. License
679
+
680
+ Both the code repository and the model weights are released under the [Kimi K3 License](https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE).
681
+
682
+ ---
683
+
684
+ ## 8. Contact Us
685
+
686
+ If you have any questions, please reach out at [support@moonshot.ai](mailto:support@moonshot.ai).