Hugging Face's logo

trymirai
/
Qwen3.6-27B-L

Text Generation
Safetensors
uzu
mirai
8-bit precision
Model card Files
xet
 Community
Roman Knyazhitskiy commited on
Commit
3d49ffc
·
1 Parent(s): f4d3b88

Add Mirai quantized weights and model card

Browse files
Files changed (5) hide show
  1. .gitattributes +1 -0
  2. README.md +1067 -0
  3. config.json +0 -0
  4. model.safetensors +3 -0
  5. tokenizer.json +3 -0
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -1,3 +1,1070 @@
1
  ---
 
 
 
 
 
2
  license: apache-2.0
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ tags:
3
+ - uzu
4
+ - mirai
5
+ - 8-bit
6
+ library_name: uzu
7
  license: apache-2.0
8
+ license_link: https://huggingface.co/Qwen/Qwen3.6-27B/blob/main/LICENSE
9
+ pipeline_tag: text-generation
10
+ base_model:
11
+ - Qwen/Qwen3.6-27B
12
+ base_model_relation: quantized
13
  ---
14
+
15
+ <div style="display:flex;align-items:center;justify-content:space-between;gap:20px;flex-wrap:wrap">
16
+ <div style="flex:1;min-width:260px">
17
+ <h1 style="margin:0 0 12px">Mirai's Qwen3.6-27B Large Quantization</h1>
18
+ <p style="margin:0">A higher-precision 8-bit quantization of Qwen3.6-27B prepared for efficient local inference on Apple silicon.</p>
19
+ </div>
20
+ <img src="https://assets.trymirai.com/images/logo/ml_small_logo.svg" alt="Mirai Labs" width="64" height="64" style="width:80px;height:80px;object-fit:contain;flex:0 0 auto">
21
+ </div>
22
+
23
+ ### Quickstart
24
+
25
+ If you are on macOS, the easiest way is to install the `mirai` Homebrew package and then run the CLI:
26
+
27
+ ```sh
28
+ brew install mirai
29
+ mirai --model trymirai/Qwen3.6-27B-L
30
+ ```
31
+
32
+ Currently only Apple silicon inference is supported. If you want to build things from source, read this [overview](https://github.com/trymirai/uzu/blob/how-to/docs/how-to-run-uzu.md).
33
+
34
+ <div style="display:flex;flex-wrap:wrap;gap:12px;align-items:center;margin-top:32px">
35
+ <a href="https://docs.trymirai.com/" style="display:inline-flex;align-items:center;justify-content:space-between;width:220px;min-height:53px;padding:0 20px;box-sizing:border-box;border:1px solid #cccccc;border-radius:0;background:#ffffff;color:#3d3d3d;text-decoration:none;font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Inter,sans-serif;font-size:15px;font-weight:500;line-height:1.5">
36
+ <span>API Docs</span>
37
+ <span aria-hidden="true" style="font-size:16px;line-height:1">&#8594;</span>
38
+ </a>
39
+ </div>
40
+
41
+ ---
42
+
43
+ ### Method
44
+
45
+ Mirai Large uses 8-bit symmetric integer quantization with bfloat16 scales and group size 64. Block-diagonal Random Hadamard Transforms are used to reduce activation and weight outliers. The checkpoint was prepared with post-training quantization.
46
+
47
+ In quantization and model-size tier, this checkpoint is broadly comparable to Unsloth's [`UD-Q8_K_XL`](https://huggingface.co/unsloth/Qwen3.6-27B-GGUF/blob/main/Qwen3.6-27B-UD-Q8_K_XL.gguf) GGUF. The formats and quantization recipes differ, so this comparison does not imply byte compatibility or identical benchmark quality.
48
+
49
+ ### Citation
50
+
51
+ If you find our work helpful, feel free to give us a cite.
52
+
53
+ ```bibtex
54
+ @misc{mirai-quant,
55
+ title = {{Mirai Quantization}: Redefining the speed-quality frontier for local LLMs on Apple silicon},
56
+ author = {Artur Chakhvadze and Ryan Mathieu and Roman Knyazhitskiy and Nikolai Voinilenko and Chen-Chen Yeh and Artur Mullakhmetov and Eugene Bokhan and others},
57
+ note = {In collaboration with others at Mirai Labs},
58
+ month = {June},
59
+ year = {2026},
60
+ url = {https://trymirai.com/blog/quantization}
61
+ }
62
+ ```
63
+
64
+ ---
65
+
66
+ The following is the model card of the original Qwen model.
67
+
68
+ # Qwen3.6-27B
69
+
70
+ <img width="400px" src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.6/logo.png">
71
+
72
+ [![Qwen Chat](https://img.shields.io/badge/%F0%9F%92%9C%EF%B8%8F%20Qwen%20Chat%20-536af5)](https://chat.qwen.ai)
73
+
74
+ > [!Note]
75
+ > This repository contains model weights and configuration files for the post-trained model in the Hugging Face Transformers format.
76
+ >
77
+ > These artifacts are compatible with Hugging Face Transformers, vLLM, SGLang, KTransformers, etc.
78
+
79
+ Following the February release of the Qwen3.5 series, we're pleased to share the first open-weight variant of Qwen3.6. Built on direct feedback from the community, Qwen3.6 prioritizes stability and real-world utility, offering developers a more intuitive, responsive, and genuinely productive coding experience.
80
+
81
+ ## Qwen3.6 Highlights
82
+
83
+ This release delivers substantial upgrades, particularly in
84
+
85
+ - **Agentic Coding:** the model now handles frontend workflows and repository-level reasoning with greater fluency and precision.
86
+ - **Thinking Preservation:** we've introduced a new option to retain reasoning context from historical messages, streamlining iterative development and reducing overhead.
87
+
88
+ ![Benchmark Results](https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3.6/Figures/qwen3.6_27b_score.png)
89
+
90
+ For more details, please refer to our blog post [Qwen3.6-27B](https://qwen.ai/blog?id=qwen3.6-27b).
91
+
92
+ ## Model Overview
93
+
94
+ - Type: Causal Language Model with Vision Encoder
95
+ - Training Stage: Pre-training & Post-training
96
+ - Language Model
97
+ - Number of Parameters: 27B
98
+ - Hidden Dimension: 5120
99
+ - Token Embedding: 248320 (Padded)
100
+ - Number of Layers: 64
101
+ - Hidden Layout: 16 × (3 × (Gated DeltaNet → FFN) → 1 × (Gated Attention → FFN))
102
+ - Gated DeltaNet:
103
+ - Number of Linear Attention Heads: 48 for V and 16 for QK
104
+ - Head Dimension: 128
105
+ - Gated Attention:
106
+ - Number of Attention Heads: 24 for Q and 4 for KV
107
+ - Head Dimension: 256
108
+ - Rotary Position Embedding Dimension: 64
109
+ - Feed Forward Network:
110
+ - Intermediate Dimension: 17408
111
+ - LM Output: 248320 (Padded)
112
+ - MTP: trained with multi-steps
113
+ - Context Length: 262,144 natively and extensible up to 1,010,000 tokens.
114
+
115
+
116
+ ## Benchmark Results
117
+
118
+ ### Language
119
+
120
+ <div style="font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif;max-width:1000px;margin:0 auto;padding:16px 0">
121
+ <table style="width:100%;border-collapse:collapse;font-size:13px">
122
+ <thead><tr>
123
+ <th style="padding:10px 7px;text-align:left;font-weight:600;border-bottom:2px solid #7c3aed;color:#7c3aed"></th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.5-27B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.5-397B-A17B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Gemma4-31B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Claude 4.5 Opus</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.6-35B-A3B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.6-27B</th></tr></thead>
124
+ <tbody>
125
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Coding Agent</td></tr>
126
+ <tr>
127
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SWE-bench Verified</td>
128
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.0</td>
129
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.2</td>
130
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.0</td>
131
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.9</td>
132
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">73.4</td>
133
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.2</td>
134
+ </tr>
135
+ <tr>
136
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SWE-bench Pro</td>
137
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.2</td>
138
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">50.9</td>
139
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">35.7</td>
140
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">57.1</td>
141
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">49.5</td>
142
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">53.5</td>
143
+ </tr>
144
+ <tr>
145
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SWE-bench Multilingual</td>
146
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">69.3</td>
147
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">69.3</td>
148
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.7</td>
149
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.5</td>
150
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.2</td>
151
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">71.3</td>
152
+ </tr>
153
+ <tr>
154
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Terminal-Bench 2.0</td>
155
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">41.6</td>
156
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.5</td>
157
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">42.9</td>
158
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">59.3</td>
159
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.5</td>
160
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">59.3</td>
161
+ </tr>
162
+ <tr>
163
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SkillsBench <sub><small>Avg5</small></sub></td>
164
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">27.2</td>
165
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">30.0</td>
166
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">23.6</td>
167
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">45.3</td>
168
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">28.7</td>
169
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">48.2</td>
170
+ </tr>
171
+ <tr>
172
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">QwenWebBench</td>
173
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1068</td>
174
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1186</td>
175
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1197</td>
176
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1536</td>
177
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1397</td>
178
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1487</td>
179
+ </tr>
180
+ <tr>
181
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">NL2Repo</td>
182
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">27.3</td>
183
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">32.2</td>
184
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">15.5</td>
185
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">43.2</td>
186
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">29.4</td>
187
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">36.2</td>
188
+ </tr>
189
+ <tr>
190
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Claw-Eval <sub><small>Avg</small></sub></td>
191
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.3</td>
192
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.7</td>
193
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">48.5</td>
194
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.6</td>
195
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">68.7</td>
196
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">72.4</td>
197
+ </tr>
198
+ <tr>
199
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Claw-Eval <sub><small>Pass^3</small></sub></td>
200
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">46.2</td>
201
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">48.1</td>
202
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">25.0</td>
203
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">59.6</td>
204
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">50.0</td>
205
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">60.6</td>
206
+ </tr>
207
+ <tr>
208
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">QwenClawBench</td>
209
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.2</td>
210
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.8</td>
211
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">41.7</td>
212
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.3</td>
213
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.6</td>
214
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">53.4</td>
215
+ </tr>
216
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Knowledge</td></tr>
217
+ <tr>
218
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMLU-Pro</td>
219
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.1</td>
220
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.8</td>
221
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.2</td>
222
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.5</td>
223
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.2</td>
224
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.2</td>
225
+ </tr>
226
+ <tr>
227
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMLU-Redux</td>
228
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.2</td>
229
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">94.9</td>
230
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.7</td>
231
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">95.6</td>
232
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.3</td>
233
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.5</td>
234
+ </tr>
235
+ <tr>
236
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SuperGPQA</td>
237
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">65.6</td>
238
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.4</td>
239
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">65.7</td>
240
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.6</td>
241
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.7</td>
242
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">66.0</td>
243
+ </tr>
244
+ <tr>
245
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">C-Eval</td>
246
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.5</td>
247
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.0</td>
248
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.6</td>
249
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.2</td>
250
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.0</td>
251
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">91.4</td>
252
+ </tr>
253
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">STEM & Reasoning</td></tr>
254
+ <tr>
255
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">GPQA Diamond</td>
256
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.5</td>
257
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">88.4</td>
258
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.3</td>
259
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.0</td>
260
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.0</td>
261
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.8</td>
262
+ </tr>
263
+ <tr>
264
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HLE</td>
265
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">24.3</td>
266
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">28.7</td>
267
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">19.5</td>
268
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">30.8</td>
269
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">21.4</td>
270
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">24.0</td>
271
+ </tr>
272
+ <tr>
273
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">LiveCodeBench v6</td>
274
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.7</td>
275
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.6</td>
276
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.0</td>
277
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.8</td>
278
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.4</td>
279
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.9</td>
280
+ </tr>
281
+ <tr>
282
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HMMT Feb 25</td>
283
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.0</td>
284
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">94.8</td>
285
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">88.7</td>
286
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.9</td>
287
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.7</td>
288
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.8</td>
289
+ </tr>
290
+ <tr>
291
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HMMT Nov 25</td>
292
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.8</td>
293
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.7</td>
294
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.5</td>
295
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.3</td>
296
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.1</td>
297
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.7</td>
298
+ </tr>
299
+ <tr>
300
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HMMT Feb 26</td>
301
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.3</td>
302
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.9</td>
303
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.2</td>
304
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.3</td>
305
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.6</td>
306
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.3</td>
307
+ </tr>
308
+ <tr>
309
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">IMOAnswerBench</td>
310
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.9</td>
311
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.9</td>
312
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.5</td>
313
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.0</td>
314
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">78.9</td>
315
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.8</td>
316
+ </tr>
317
+ <tr>
318
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">AIME26</td>
319
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.6</td>
320
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.3</td>
321
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.2</td>
322
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">95.1</td>
323
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.7</td>
324
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">94.1</td>
325
+ </tr>
326
+ </tbody>
327
+ </table>
328
+
329
+ <p style="margin-top:12px;font-size:10px;opacity:0.7">
330
+ * SWE-Bench Series: Internal agent scaffold (bash + file-edit tools); temp=1.0, top_p=0.95, 200K context window. We correct some problematic tasks in the public set of SWE-bench Pro and evaluate all baselines on the refined benchmark.<br/>
331
+ * Terminal-Bench 2.0: Harbor/Terminus-2 harness; 3h timeout, 32 CPU/48 GB RAM; temp=1.0, top_p=0.95, top_k=20, max_tokens=80K, 256K ctx; avg of 5 runs.<br/>
332
+ * SkillsBench: Evaluated via OpenCode on 78 tasks (self-contained subset, excluding API-dependent tasks); avg of 5 runs.<br/>
333
+ * NL2Repo: Others are evaluated via Claude Code (temp=1.0, top_p=0.95, max_turns=900).<br/>
334
+ * QwenClawBench: A real-user-distribution Claw agent benchmark; temp=0.6, 256K ctx.<br/>
335
+ * QwenWebBench: An internal front-end code generation benchmark; bilingual (EN/CN), 7 categories (Web Design, Web Apps, Games, SVG, Data Visualization, Animation, and 3D); auto-render + multimodal judge (code/visual correctness); BT/Elo rating system.<br/>
336
+ * AIME 26: We use the full AIME 2026 (I & II), where the scores may differ from Qwen 3.5 notes.
337
+ </p>
338
+
339
+ </div>
340
+
341
+
342
+ ### Vision Language
343
+
344
+ <div style="font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif;max-width:1000px;margin:0 auto;padding:16px 0">
345
+ <table style="width:100%;border-collapse:collapse;font-size:13px">
346
+ <thead><tr>
347
+ <th style="padding:10px 7px;text-align:left;font-weight:600;border-bottom:2px solid #7c3aed;color:#7c3aed"></th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.5-27B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.5-397B-A17B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Gemma4-31B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Claude 4.5 Opus</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.6-35B-A3B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.6-27B</th></tr></thead>
348
+ <tbody>
349
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">STEM & Puzzle</td></tr>
350
+ <tr>
351
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMMU</td>
352
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.3</td>
353
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.0</td>
354
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.4</td>
355
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.7</td>
356
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.7</td>
357
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.9</td>
358
+ </tr>
359
+ <tr>
360
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMMU-Pro</td>
361
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.0</td>
362
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.0</td>
363
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.9</td>
364
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.6</td>
365
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.3</td>
366
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.8</td>
367
+ </tr>
368
+ <tr>
369
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MathVista <sub><small>mini</small></sub></td>
370
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.8</td>
371
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
372
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.3</td>
373
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
374
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.4</td>
375
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.4</td>
376
+ </tr>
377
+ <tr>
378
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">DynaMath</td>
379
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.7</td>
380
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.3</td>
381
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.5</td>
382
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.7</td>
383
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.8</td>
384
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.6</td>
385
+ </tr>
386
+ <tr>
387
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">VlmsAreBlind</td>
388
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">96.9</td>
389
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
390
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.2</td>
391
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
392
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">96.6</td>
393
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">97.0</td>
394
+ </tr>
395
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">General VQA</td></tr>
396
+ <tr>
397
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">RealWorldQA</td>
398
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.7</td>
399
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.9</td>
400
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">72.3</td>
401
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.0</td>
402
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.3</td>
403
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.1</td>
404
+ </tr>
405
+ <tr>
406
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMStar</td>
407
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.0</td>
408
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.8</td>
409
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.3</td>
410
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">73.2</td>
411
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.7</td>
412
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.4</td>
413
+ </tr>
414
+ <tr>
415
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMBench<sub><small>EN-DEV-v1.1</small></sub></td>
416
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.6</td>
417
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
418
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.9</td>
419
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
420
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.8</td>
421
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.3</td>
422
+ </tr>
423
+ <tr>
424
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SimpleVQA</td>
425
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">56.0</td>
426
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.1</td>
427
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.9</td>
428
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">65.7</td>
429
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">58.9</td>
430
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">56.1</td>
431
+ </tr>
432
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Document Understanding</td></tr>
433
+ <tr>
434
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">CharXiv <sub><small>RQ</small></sub></td>
435
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.5</td>
436
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.8</td>
437
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.9</td>
438
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">68.5</td>
439
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">78.0</td>
440
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">78.4</td>
441
+ </tr>
442
+ <tr>
443
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">CC-OCR</td>
444
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.0</td>
445
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.0</td>
446
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.7</td>
447
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.9</td>
448
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.9</td>
449
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.2</td>
450
+ </tr>
451
+ <tr>
452
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">OCRBench</td>
453
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.4</td>
454
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
455
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.1</td>
456
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
457
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.0</td>
458
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.4</td>
459
+ </tr>
460
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Spatial Intelligence</td></tr>
461
+ <tr>
462
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">ERQA</td>
463
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">60.5</td>
464
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.5</td>
465
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">57.5</td>
466
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">46.8</td>
467
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">61.8</td>
468
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">62.5</td>
469
+ </tr>
470
+ <tr>
471
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">CountBench</td>
472
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">97.8</td>
473
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">97.2</td>
474
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">96.1</td>
475
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.6</td>
476
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">96.1</td>
477
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">97.8</td>
478
+ </tr>
479
+ <tr>
480
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">RefCOCO <sub><small>avg</small></sub></td>
481
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.9</td>
482
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.3</td>
483
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
484
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
485
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.0</td>
486
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.5</td>
487
+ </tr>
488
+ <tr>
489
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">EmbSpatialBench</td>
490
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.5</td>
491
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
492
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
493
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
494
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.3</td>
495
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.6</td>
496
+ </tr>
497
+ <tr>
498
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">RefSpatialBench</td>
499
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.7</td>
500
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
501
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">4.7</td>
502
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
503
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.3</td>
504
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.0</td>
505
+ </tr>
506
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Video Understanding</td></tr>
507
+ <tr>
508
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">VideoMME<sub><small>(w sub.)</sub></small></td>
509
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.0</td>
510
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.5</td>
511
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
512
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.7</td>
513
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.6</td>
514
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.7</td>
515
+ </tr>
516
+ <tr>
517
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">VideoMMMU</td>
518
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.3</td>
519
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.7</td>
520
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.6</td>
521
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.4</td>
522
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.7</td>
523
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.4</td>
524
+ </tr>
525
+ <tr>
526
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MLVU</td>
527
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.9</td>
528
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.7</td>
529
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
530
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.7</td>
531
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.2</td>
532
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.6</td>
533
+ </tr>
534
+ <tr>
535
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MVBench</td>
536
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.6</td>
537
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.6</td>
538
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
539
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.2</td>
540
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.6</td>
541
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.5</td>
542
+ </tr>
543
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Visual Agent</td></tr>
544
+ <tr>
545
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">V*</td>
546
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.7</td>
547
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">95.8</td>
548
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
549
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.0</td>
550
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.1</td>
551
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">94.7</td>
552
+ </tr>
553
+ <tr>
554
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">AndroidWorld</td>
555
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.2</td>
556
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
557
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
558
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
559
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
560
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.3</td>
561
+ </tr>
562
+ </tbody>
563
+ </table>
564
+
565
+ <p style="margin-top:12px;font-size:10px;opacity:0.7">
566
+ * Empty cells (--) indicate scores not yet available or not applicable.
567
+ </p>
568
+
569
+ </div>
570
+
571
+
572
+ ## Quickstart
573
+
574
+ For streamlined integration, we recommend using Qwen3.6 via APIs. Below is a guide to use Qwen3.6 via OpenAI-compatible API.
575
+
576
+ ### Serving Qwen3.6
577
+
578
+ Qwen3.6 can be served via APIs with popular inference frameworks.
579
+ In the following, we show example commands to launch OpenAI-Compatible API servers for Qwen3.6 models.
580
+
581
+ > [!Important]
582
+ > Inference efficiency and throughput vary significantly across frameworks.
583
+ > We recommend using the latest framework versions to ensure optimal performance and compatibility.
584
+ > For production workloads or high-throughput scenarios, dedicated serving engines such as SGLang, KTransformers or vLLM are strongly recommended.
585
+
586
+ > [!Important]
587
+ > The model has a default context length of 262,144 tokens.
588
+ > If you encounter out-of-memory (OOM) errors, consider reducing the context window.
589
+ > However, because Qwen3.6 leverages extended context for complex tasks, we advise maintaining a context length of at least 128K tokens to preserve thinking capabilities.
590
+
591
+ #### SGLang
592
+
593
+ [SGLang](https://github.com/sgl-project/sglang) is a fast serving framework for large language models and vision language models.
594
+ `sglang>=0.5.10` is recommended for Qwen3.6, which can be installed using the following command in a fresh environment:
595
+ ```shell
596
+ uv pip install sglang[all]
597
+ ```
598
+ See [its documentation](https://docs.sglang.ai/get_started/install.html) for more details.
599
+
600
+ The following will create API endpoints at `http://localhost:8000/v1`:
601
+
602
+ - **Standard Version**: The following command can be used to create an API endpoint with maximum context length 262,144 tokens using tensor parallel on 8 GPUs.
603
+
604
+ ```shell
605
+ python -m sglang.launch_server --model-path Qwen/Qwen3.6-27B --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3
606
+ ```
607
+
608
+ - **Tool Use**: To support tool use, you can use the following command.
609
+
610
+ ```shell
611
+ python -m sglang.launch_server --model-path Qwen/Qwen3.6-27B --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3 --tool-call-parser qwen3_coder
612
+ ```
613
+
614
+ - **Multi-Token Prediction (MTP)**: The following command is recommended for MTP:
615
+
616
+ ```shell
617
+ python -m sglang.launch_server --model-path Qwen/Qwen3.6-27B --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3 --speculative-algo NEXTN --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4
618
+ ```
619
+
620
+ For detailed deployment guide, see the [SGLang Qwen3.5 Cookbook](https://lmsysorg.mintlify.app/cookbook/llm/Qwen/Qwen3.5).
621
+
622
+ #### vLLM
623
+
624
+ [vLLM](https://github.com/vllm-project/vllm) is a high-throughput and memory-efficient inference and serving engine for LLMs.
625
+ `vllm>=0.19.0` is recommended for Qwen3.6, which can be installed using the following command in a fresh environment:
626
+ ```shell
627
+ uv pip install vllm --torch-backend=auto
628
+ ```
629
+ See [its documentation](https://docs.vllm.ai/en/stable/getting_started/installation/index.html) for more details.
630
+
631
+
632
+ The following will create API endpoints at `http://localhost:8000/v1`:
633
+
634
+ - **Standard Version**: The following command can be used to create an API endpoint with maximum context length 262,144 tokens using tensor parallel on 8 GPUs.
635
+
636
+ ```shell
637
+ vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3
638
+ ```
639
+
640
+ - **Tool Call**: To support tool use, you can use the following command.
641
+
642
+ ```shell
643
+ vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder
644
+ ```
645
+
646
+ - **Multi-Token Prediction (MTP)**: The following command is recommended for MTP:
647
+
648
+ ```shell
649
+ vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --speculative-config '{"method":"qwen3_next_mtp","num_speculative_tokens":2}'
650
+ ```
651
+
652
+ - **Text-Only**: The following command skips the vision encoder and multimodal profiling to free up memory for additional KV cache:
653
+
654
+ ```shell
655
+ vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --language-model-only
656
+ ```
657
+
658
+ For detailed deployment guide, see the [vLLM Qwen3.5 Recipe](https://docs.vllm.ai/projects/recipes/en/latest/Qwen/Qwen3.5.html).
659
+
660
+ #### KTransformers
661
+
662
+ [KTransformers](https://github.com/kvcache-ai/ktransformers) is a flexible framework for experiencing cutting-edge LLM inference optimizations with CPU-GPU heterogeneous computing.
663
+ For running Qwen3.6 with KTransformers, see the [KTransformers Deployment Guide](https://github.com/kvcache-ai/ktransformers/blob/main/doc/en/Qwen3.5.md).
664
+
665
+ #### Hugging Face Transformers
666
+
667
+ Hugging Face Transformers contains a _lightweight_ server which can be used for quick testing and moderate load deployment.
668
+ The latest `transformers` is required for Qwen3.6:
669
+ ```shell
670
+ pip install "transformers[serving]"
671
+ ```
672
+ See [its documentation](https://huggingface.co/docs/transformers/main/serving) for more details. Please also make sure torchvision and pillow are installed.
673
+
674
+ Then, run `transformers serve` to launch a server with API endpoints at `http://localhost:8000/v1`; it will place the model on accelerators if available:
675
+ ```shell
676
+ transformers serve Qwen/Qwen3.6-27B --port 8000 --continuous-batching
677
+ ```
678
+
679
+ ### Using Qwen3.6 via the Chat Completions API
680
+
681
+ The chat completions API is accessible via standard HTTP requests or OpenAI SDKs.
682
+ Here, we show examples using the OpenAI Python SDK.
683
+
684
+ Before starting, make sure it is installed and the API key and the API base URL is configured, e.g.:
685
+ ```shell
686
+ pip install -U openai
687
+
688
+ # Set the following accordingly
689
+ export OPENAI_BASE_URL="http://localhost:8000/v1"
690
+ export OPENAI_API_KEY="EMPTY"
691
+ ```
692
+
693
+ > [!Tip]
694
+ > We recommend using the following set of sampling parameters for generation
695
+ > - Thinking mode for general tasks: `temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0`
696
+ > - Thinking mode for precise coding tasks (e.g. WebDev): `temperature=0.6, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0`
697
+ > - Instruct (or non-thinking) mode: `temperature=0.7, top_p=0.80, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0`
698
+ >
699
+ > Please note that the support for sampling parameters varies according to inference frameworks.
700
+
701
+ > [!Important]
702
+ > Qwen3.6 models operate in thinking mode by default, generating thinking content signified by `<think>\n...</think>\n\n` before producing the final responses.
703
+ > To disable thinking content and obtain direct response, refer to the examples [here](#instruct-or-non-thinking-mode).
704
+
705
+
706
+ #### Text-Only Input
707
+
708
+ ```python
709
+ from openai import OpenAI
710
+ # Configured by environment variables
711
+ client = OpenAI()
712
+
713
+ messages = [
714
+ {"role": "user", "content": "Type \"I love Qwen3.6\" backwards"},
715
+ ]
716
+
717
+ chat_response = client.chat.completions.create(
718
+ model="Qwen/Qwen3.6-27B",
719
+ messages=messages,
720
+ max_tokens=81920,
721
+ temperature=1.0,
722
+ top_p=0.95,
723
+ presence_penalty=0.0,
724
+ extra_body={
725
+ "top_k": 20,
726
+ },
727
+ )
728
+ print("Chat response:", chat_response)
729
+ ```
730
+
731
+
732
+ #### Image Input
733
+
734
+ ```python
735
+ from openai import OpenAI
736
+ # Configured by environment variables
737
+ client = OpenAI()
738
+
739
+ messages = [
740
+ {
741
+ "role": "user",
742
+ "content": [
743
+ {
744
+ "type": "image_url",
745
+ "image_url": {
746
+ "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/CI_Demo/mathv-1327.jpg"
747
+ }
748
+ },
749
+ {
750
+ "type": "text",
751
+ "text": "The centres of the four illustrated circles are in the corners of the square. The two big circles touch each other and also the two little circles. With which factor do you have to multiply the radii of the little circles to obtain the radius of the big circles?\nChoices:\n(A) $\\frac{2}{9}$\n(B) $\\sqrt{5}$\n(C) $0.8 \\cdot \\pi$\n(D) 2.5\n(E) $1+\\sqrt{2}$"
752
+ }
753
+ ]
754
+ }
755
+ ]
756
+
757
+ chat_response = client.chat.completions.create(
758
+ model="Qwen/Qwen3.6-27B",
759
+ messages=messages,
760
+ max_tokens=81920,
761
+ temperature=1.0,
762
+ top_p=0.95,
763
+ presence_penalty=0.0,
764
+ extra_body={
765
+ "top_k": 20,
766
+ },
767
+ )
768
+ print("Chat response:", chat_response)
769
+ ```
770
+
771
+ #### Video Input
772
+
773
+ ```python
774
+ from openai import OpenAI
775
+ # Configured by environment variables
776
+ client = OpenAI()
777
+
778
+ messages = [
779
+ {
780
+ "role": "user",
781
+ "content": [
782
+ {
783
+ "type": "video_url",
784
+ "video_url": {
785
+ "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/video/N1cdUjctpG8.mp4"
786
+ }
787
+ },
788
+ {
789
+ "type": "text",
790
+ "text": "How many porcelain jars were discovered in the niches located in the primary chamber of the tomb?"
791
+ }
792
+ ]
793
+ }
794
+ ]
795
+
796
+ # When vLLM is launched with `--media-io-kwargs '{"video": {"num_frames": -1}}'`,
797
+ # video frame sampling can be configured via `extra_body` (e.g., by setting `fps`).
798
+ # This feature is currently supported only in vLLM.
799
+ #
800
+ # By default, `fps=2` and `do_sample_frames=True`.
801
+ # With `do_sample_frames=True`, you can customize the `fps` value to set your desired video sampling rate.
802
+ chat_response = client.chat.completions.create(
803
+ model="Qwen/Qwen3.6-27B",
804
+ messages=messages,
805
+ max_tokens=81920,
806
+ temperature=1.0,
807
+ top_p=0.95,
808
+ presence_penalty=0.0,
809
+ extra_body={
810
+ "top_k": 20,
811
+ "mm_processor_kwargs": {"fps": 2, "do_sample_frames": True},
812
+ },
813
+ )
814
+
815
+ print("Chat response:", chat_response)
816
+ ```
817
+
818
+
819
+ #### Instruct (or Non-Thinking) Mode
820
+
821
+ > [!Important]
822
+ > Qwen3.6 does not officially support the soft switch of Qwen3, i.e., `/think` and `/nothink`.
823
+
824
+ Qwen3.6 will think by default before response.
825
+ You can obtain direct response from the model without thinking by configuring the API parameters.
826
+ For example,
827
+ ```python
828
+ from openai import OpenAI
829
+ # Configured by environment variables
830
+ client = OpenAI()
831
+
832
+ messages = [
833
+ {
834
+ "role": "user",
835
+ "content": [
836
+ {
837
+ "type": "image_url",
838
+ "image_url": {
839
+ "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.6/demo/RealWorld/RealWorld-04.png"
840
+ }
841
+ },
842
+ {
843
+ "type": "text",
844
+ "text": "Where is this?"
845
+ }
846
+ ]
847
+ }
848
+ ]
849
+
850
+ chat_response = client.chat.completions.create(
851
+ model="Qwen/Qwen3.6-27B",
852
+ messages=messages,
853
+ max_tokens=32768,
854
+ temperature=0.7,
855
+ top_p=0.8,
856
+ presence_penalty=1.5,
857
+ extra_body={
858
+ "top_k": 20,
859
+ "chat_template_kwargs": {"enable_thinking": False},
860
+ },
861
+ )
862
+ print("Chat response:", chat_response)
863
+ ```
864
+
865
+ > [!Note]
866
+ > If you are using APIs from Alibaba Cloud Model Studio, in addition to changing `model`, please use `"enable_thinking": False` instead of `"chat_template_kwargs": {"enable_thinking": False}`.
867
+
868
+ #### Preserve Thinking
869
+
870
+ By default, only the thinking blocks generated in handling the latest user message is retained, resulting in a pattern commonly as interleaved thinking.
871
+ Qwen3.6 has been additionally trained to preserve and leverage thinking traces from historical messages.
872
+ You can enable this behavior by setting the `preserve_thinking` option:
873
+ ```python
874
+ from openai import OpenAI
875
+ # Configured by environment variables
876
+ client = OpenAI()
877
+
878
+ messages = [...]
879
+
880
+ chat_response = client.chat.completions.create(
881
+ model="Qwen/Qwen3.6-27B",
882
+ messages=messages,
883
+ max_tokens=32768,
884
+ temperature=0.6,
885
+ top_p=0.95,
886
+ presence_penalty=0.0,
887
+ extra_body={
888
+ "top_k": 20,
889
+ "chat_template_kwargs": {"preserve_thinking": True},
890
+ },
891
+ )
892
+ print("Chat response:", chat_response)
893
+ ```
894
+
895
+ > [!Note]
896
+ > If you are using APIs from Alibaba Cloud Model Studio, in addition to changing `model`, please use `"preserve_thinking": True` instead of `"chat_template_kwargs": {"preserve_thinking": False}`.
897
+
898
+
899
+ This capability is particularly beneficial for agent scenarios, where maintaining full reasoning context can enhance decision consistency and, in many cases, reduce overall token consumption by minimizing redundant reasoning. Additionally, it can improve KV cache utilization, optimizing inference efficiency in both thinking and non-thinking modes.
900
+
901
+
902
+ ## Agentic Usage
903
+
904
+ Qwen3.6 excels in tool calling capabilities.
905
+
906
+ ### Qwen-Agent
907
+
908
+ We recommend using [Qwen-Agent](https://github.com/QwenLM/Qwen-Agent) to quickly build Agent applications with Qwen3.6.
909
+
910
+ To define the available tools, you can use the MCP configuration file, use the integrated tool of Qwen-Agent, or integrate other tools by yourself.
911
+ ```python
912
+ import os
913
+ from qwen_agent.agents import Assistant
914
+
915
+ # Define LLM
916
+ # Using Alibaba Cloud Model Studio
917
+ llm_cfg = {
918
+ # Use the OpenAI-compatible model service provided by DashScope:
919
+ 'model': 'qwen3.6-27b',
920
+ 'model_type': 'qwenvl_oai',
921
+ 'model_server': 'https://dashscope.aliyuncs.com/compatible-mode/v1',
922
+ 'api_key': os.getenv('DASHSCOPE_API_KEY'),
923
+
924
+ 'generate_cfg': {
925
+ 'use_raw_api': True,
926
+ # When using Dash Scope OAI API, pass the parameter of whether to enable thinking mode in this way
927
+ 'extra_body': {
928
+ 'enable_thinking': True,
929
+ 'preserve_thinking': True,
930
+ },
931
+ },
932
+ }
933
+
934
+ # Using OpenAI-compatible API endpoint.
935
+ # functionality of the deployment frameworks and let Qwen-Agent automate the related operations.
936
+ #
937
+ # llm_cfg = {
938
+ # # Use your own model service compatible with OpenAI API by vLLM/SGLang:
939
+ # 'model': 'Qwen/Qwen3.6-27B',
940
+ # 'model_type': 'qwenvl_oai',
941
+ # 'model_server': 'http://localhost:8000/v1', # api_base
942
+ # 'api_key': 'EMPTY',
943
+ #
944
+ # 'generate_cfg': {
945
+ # 'use_raw_api': True,
946
+ # # When using vLLM/SGLang OAI API, pass the parameter of whether to enable thinking mode in this way
947
+ # 'extra_body': {
948
+ # 'chat_template_kwargs': {'enable_thinking': True, 'preserve_thinking': True}
949
+ # },
950
+ # },
951
+ # }
952
+
953
+ # Define Tools
954
+ tools = [
955
+ {'mcpServers': { # You can specify the MCP configuration file
956
+ "filesystem": {
957
+ "command": "npx",
958
+ "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/xxxx/Desktop"]
959
+ }
960
+ }
961
+ }
962
+ ]
963
+
964
+ # Define Agent
965
+ bot = Assistant(llm=llm_cfg, function_list=tools)
966
+
967
+ # Streaming generation
968
+ messages = [{'role': 'user', 'content': 'Help me organize my desktop.'}]
969
+ for responses in bot.run(messages=messages):
970
+ pass
971
+ print(responses)
972
+
973
+ # Streaming generation
974
+ messages = [{'role': 'user', 'content': 'Develop a dog website and save it on the desktop'}]
975
+ for responses in bot.run(messages=messages):
976
+ pass
977
+ print(responses)
978
+ ```
979
+
980
+ ### Qwen Code
981
+
982
+
983
+ [Qwen Code](https://github.com/QwenLM/qwen-code) is an open-source AI agent for the terminal, optimized for Qwen models. It helps you understand large codebases, automate tedious work, and ship faster.
984
+
985
+ For more information, please refer to [Qwen Code](https://qwenlm.github.io/qwen-code-docs/).
986
+
987
+ ## Processing Ultra-Long Texts
988
+
989
+ Qwen3.6 natively supports context lengths of up to 262,144 tokens.
990
+ For long-horizon tasks where the total length (including both input and output) exceeds this limit, we recommend using RoPE scaling techniques to handle long texts effectively., e.g., YaRN.
991
+
992
+ YaRN is currently supported by several inference frameworks, e.g., `transformers`, `vllm`, `ktransformers` and `sglang`.
993
+ In general, there are two approaches to enabling YaRN for supported frameworks:
994
+
995
+ - Modifying the model configuration file:
996
+ In the `config.json` file, change the `rope_parameters` fields in `text_config` to:
997
+ ```json
998
+ {
999
+ "mrope_interleaved": true,
1000
+ "mrope_section": [
1001
+ 11,
1002
+ 11,
1003
+ 10
1004
+ ],
1005
+ "rope_type": "yarn",
1006
+ "rope_theta": 10000000,
1007
+ "partial_rotary_factor": 0.25,
1008
+ "factor": 4.0,
1009
+ "original_max_position_embeddings": 262144,
1010
+ }
1011
+ ```
1012
+
1013
+ - Passing command line arguments:
1014
+
1015
+ For `vllm`, you can use
1016
+ ```shell
1017
+ VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 vllm serve ... --hf-overrides '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --max-model-len 1010000
1018
+ ```
1019
+
1020
+ For `sglang` and `ktransformers`, you can use
1021
+ ```shell
1022
+ SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 python -m sglang.launch_server ... --json-model-override-args '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --context-length 1010000
1023
+ ```
1024
+
1025
+ > [!NOTE]
1026
+ > All the notable open-source frameworks implement static YaRN, which means the scaling factor remains constant regardless of input length, **potentially impacting performance on shorter texts.**
1027
+ > We advise modifying the `rope_parameters` configuration only when processing long contexts is required.
1028
+ > It is also recommended to modify the `factor` as needed. For example, if the typical context length for your application is 524,288 tokens, it would be better to set `factor` as 2.0.
1029
+
1030
+ ## Best Practices
1031
+
1032
+ To achieve optimal performance, we recommend the following settings:
1033
+
1034
+ 1. **Sampling Parameters**:
1035
+ - We suggest using the following sets of sampling parameters depending on the mode and task type:
1036
+ - **Thinking mode for general tasks**:
1037
+ `temperature=1.0`, `top_p=0.95`, `top_k=20`, `min_p=0.0`, `presence_penalty=0.0`, `repetition_penalty=1.0`
1038
+ - **Thinking mode for precise coding tasks (e.g., WebDev)**:
1039
+ `temperature=0.6`, `top_p=0.95`, `top_k=20`, `min_p=0.0`, `presence_penalty=0.0`, `repetition_penalty=1.0`
1040
+ - **Instruct (or non-thinking) mode**:
1041
+ `temperature=0.7`, `top_p=0.80`, `top_k=20`, `min_p=0.0`, `presence_penalty=1.5`, `repetition_penalty=1.0`
1042
+ - For supported frameworks, you can adjust the `presence_penalty` parameter between 0 and 2 to reduce endless repetitions. However, using a higher value may occasionally result in language mixing and a slight decrease in model performance.
1043
+
1044
+ 2. **Adequate Output Length**: We recommend using an output length of 32,768 tokens for most queries. For benchmarking on highly complex problems, such as those found in math and programming competitions, we suggest setting the max output length to 81,920 tokens. This provides the model with sufficient space to generate detailed and comprehensive responses, thereby enhancing its overall performance.
1045
+
1046
+ 3. **Standardize Output Format**: We recommend using prompts to standardize model outputs when benchmarking.
1047
+ - **Math Problems**: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt.
1048
+ - **Multiple-Choice Questions**: Add the following JSON structure to the prompt to standardize responses: "Please show your choice in the `answer` field with only the choice letter, e.g., `"answer": "C"`."
1049
+
1050
+ 4. **Long Video Understanding**: To optimize inference efficiency for plain text and images, the `size` parameter in the released `video_preprocessor_config.json` is conservatively configured. It is recommended to set the `longest_edge` parameter in the video_preprocessor_config file to 469,762,048 (corresponding to 224k video tokens) to enable higher frame-rate sampling for hour-scale videos and thereby achieve superior performance. For example,
1051
+ ```json
1052
+ {"longest_edge": 469762048, "shortest_edge": 4096}
1053
+ ```
1054
+
1055
+ Alternatively, override the default values via engine startup parameters. For implementation details, refer to: [vLLM](https://github.com/vllm-project/vllm/pull/34330) / [SGLang](https://github.com/sgl-project/sglang/pull/18467).
1056
+
1057
+
1058
+ ### Citation
1059
+
1060
+ If you find our work helpful, feel free to give us a cite.
1061
+
1062
+ ```bibtex
1063
+ @misc{qwen3.6-27b,
1064
+ title = {{Qwen3.6-27B}: Flagship-Level Coding in a {27B} Dense Model},
1065
+ author = {{Qwen Team}},
1066
+ month = {April},
1067
+ year = {2026},
1068
+ url = {https://qwen.ai/blog?id=qwen3.6-27b}
1069
+ }
1070
+ ```
config.json ADDED
The diff for this file is too large to render. See raw diff
 
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:28f82e20ac74caaea82971b2221e4ad841cdb1ea239a039195223370088f3896
3
+ size 27769349616
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:87a7830d63fcf43bf241c3c5242e96e62dd3fdc29224ca26fed8ea333db72de4
3
+ size 19989343