README.md · Qwen3.6-35B-A3B-FP8

1

---

2

base_model:

3

- Qwen/Qwen3.6-35B-A3B

4

frameworks:

5

- ""

6

library_name: transformers

7

license: apache-2.0

8

license_link: https://huggingface.co/Qwen/Qwen3.6-35B-A3B-FP8/blob/main/LICENSE

9

pipeline_tag: image-text-to-text

10

tasks: []

11

---

12

13

# Qwen3.6-35B-A3B-FP8

14

15

<img width="400px" src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.6/logo.png">

16

17

[![Qwen Chat](https://img.shields.io/badge/%F0%9F%92%9C%EF%B8%8F%20Qwen%20Chat%20-536af5)](https://chat.qwen.ai)

18

19

> [!Note]

20

> This repository contains FP8-quantized model weights and configuration files for the post-trained model in the Hugging Face Transformers format.

21

>

22

> These artifacts are compatible with Hugging Face Transformers, vLLM, SGLang, KTransformers, etc.

23

>

24

> The quantization method is fine-grained fp8 quantization with block size of 128, and its performance metrics are nearly identical to those of the original model.

25

26

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.

27

28

## Qwen3.6 Highlights

29

30

This release delivers substantial upgrades, particularly in

31

32

- **Agentic Coding:** the model now handles frontend workflows and repository-level reasoning with greater fluency and precision.

33

- **Thinking Preservation:** we've introduced a new option to retain reasoning context from historical messages, streamlining iterative development and reducing overhead.

34

35

![Benchmark Results](https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3.6/Figures/qwen3.6_35b_a3b_score.png)

36

37

For more details, please refer to our blog post [Qwen3.6-35B-A3B](https://qwen.ai/blog?id=qwen3.6-35b-a3b).

38

39

## Model Overview

40

41

- Type: Causal Language Model with Vision Encoder

42

- Training Stage: Pre-training & Post-training

43

- Language Model

44

- Number of Parameters: 35B in total and 3B activated

45

- Hidden Dimension: 2048

46

- Token Embedding: 248320 (Padded)

47

- Number of Layers: 40

48

- Hidden Layout: 10 × (3 × (Gated DeltaNet → MoE) → 1 × (Gated Attention → MoE))

49

- Gated DeltaNet:

50

- Number of Linear Attention Heads: 32 for V and 16 for QK

51

- Head Dimension: 128

52

- Gated Attention:

53

- Number of Attention Heads: 16 for Q and 2 for KV

54

- Head Dimension: 256

55

- Rotary Position Embedding Dimension: 64

56

- Mixture Of Experts

57

- Number of Experts: 256

58

- Number of Activated Experts: 8 Routed + 1 Shared

59

- Expert Intermediate Dimension: 512

60

- LM Output: 248320 (Padded)

61

- MTP: trained with multi-steps

62

- Context Length: 262,144 natively and extensible up to 1,010,000 tokens.

63

64

65

## Benchmark Results

66

67

### Language

68

69

<div style="font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif;max-width:1000px;margin:0 auto;padding:16px 0">

70

<table style="width:100%;border-collapse:collapse;font-size:13px">

71

<thead><tr>

72

<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;">Gemma4-31B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.5-35BA3B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Gemma4-26BA4B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.6-35BA3B</th></tr></thead>

73

<tbody>

74

<tr><td colspan="6" 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>

75

<tr>

76

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SWE-bench Verified</td>

77

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.0</td>

78

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.0</td>

79

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.0</td>

80

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">17.4</td>

81

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">73.4</td>

82

</tr>

83

<tr>

84

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SWE-bench Multilingual</td>

85

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">69.3</td>

86

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.7</td>

87

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">60.3</td>

88

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">17.3</td>

89

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.2</td>

90

</tr>

91

<tr>

92

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SWE-bench Pro</td>

93

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.2</td>

94

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">35.7</td>

95

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">44.6</td>

96

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">13.8</td>

97

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">49.5</td>

98

</tr>

99

<tr>

100

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Terminal-Bench 2.0</td>

101

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">41.6</td>

102

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">42.9</td>

103

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">40.5</td>

104

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">34.2</td>

105

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.5</td>

106

</tr>

107

<tr>

108

<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>

109

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.3</td>

110

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">48.5</td>

111

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">65.4</td>

112

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">58.8</td>

113

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">68.7</td>

114

</tr>

115

<tr>

116

<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>

117

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">46.2</td>

118

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">25.0</td>

119

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.0</td>

120

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">28.0</td>

121

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">50.0</td>

122

</tr>

123

<tr>

124

<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>

125

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">27.2</td>

126

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">23.6</td>

127

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">4.4</td>

128

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">12.3</td>

129

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">28.7</td>

130

</tr>

131

<tr>

132

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">QwenClawBench</td>

133

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.2</td>

134

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">41.7</td>

135

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">47.7</td>

136

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">38.7</td>

137

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.6</td>

138

</tr>

139

<tr>

140

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">NL2Repo</td>

141

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">27.3</td>

142

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">15.5</td>

143

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">20.5</td>

144

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">11.6</td>

145

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">29.4</td>

146

</tr>

147

<tr>

148

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">QwenWebBench</td>

149

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1068</td>

150

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1197</td>

151

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">978</td>

152

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1178</td>

153

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1397</td>

154

</tr>

155

<tr><td colspan="6" 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 Agent</td></tr>

156

<tr>

157

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">TAU3-Bench</td>

158

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">68.4</td>

159

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.5</td>

160

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">68.9</td>

161

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">59.0</td>

162

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.2</td>

163

</tr>

164

<tr>

165

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">VITA-Bench</td>

166

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">41.8</td>

167

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">43.0</td>

168

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">29.1</td>

169

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">36.9</td>

170

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">35.6</td>

171

</tr>

172

<tr>

173

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">DeepPlanning</td>

174

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">22.6</td>

175

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">24.0</td>

176

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">22.8</td>

177

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">16.2</td>

178

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">25.9</td>

179

</tr>

180

<tr>

181

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Tool Decathlon</td>

182

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">31.5</td>

183

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">21.2</td>

184

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">28.7</td>

185

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">12.0</td>

186

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">26.9</td>

187

</tr>

188

<tr>

189

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MCPMark</td>

190

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">36.3</td>

191

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">18.1</td>

192

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">27.0</td>

193

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">14.2</td>

194

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">37.0</td>

195

</tr>

196

<tr>

197

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MCP-Atlas</td>

198

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">68.4</td>

199

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">57.2</td>

200

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">62.4</td>

201

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">50.0</td>

202

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">62.8</td>

203

</tr>

204

<tr>

205

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">WideSearch</td>

206

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">66.4</td>

207

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">35.2</td>

208

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">59.1</td>

209

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">38.3</td>

210

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">60.1</td>

211

</tr>

212

<tr><td colspan="6" 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>

213

<tr>

214

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMLU-Pro</td>

215

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.1</td>

216

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.2</td>

217

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.3</td>

218

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.6</td>

219

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.2</td>

220

</tr>

221

<tr>

222

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMLU-Redux</td>

223

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.2</td>

224

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.7</td>

225

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.3</td>

226

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.7</td>

227

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.3</td>

228

</tr>

229

<tr>

230

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SuperGPQA</td>

231

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">65.6</td>

232

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">65.7</td>

233

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">63.4</td>

234

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">61.4</td>

235

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.7</td>

236

</tr>

237

<tr>

238

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">C-Eval</td>

239

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.5</td>

240

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.6</td>

241

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.2</td>

242

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.5</td>

243

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.0</td>

244

</tr>

245

<tr><td colspan="6" 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>

246

<tr>

247

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">GPQA</td>

248

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.5</td>

249

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.3</td>

250

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.2</td>

251

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.3</td>

252

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.0</td>

253

</tr>

254

<tr>

255

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HLE</td>

256

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">24.3</td>

257

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">19.5</td>

258

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">22.4</td>

259

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">8.7</td>

260

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">21.4</td>

261

</tr>

262

<tr>

263

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">LiveCodeBench v6</td>

264

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.7</td>

265

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.0</td>

266

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.6</td>

267

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.1</td>

268

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.4</td>

269

</tr>

270

<tr>

271

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HMMT Feb 25</td>

272

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.0</td>

273

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">88.7</td>

274

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.0</td>

275

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">91.7</td>

276

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.7</td>

277

</tr>

278

<tr>

279

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HMMT Nov 25</td>

280

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.8</td>

281

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.5</td>

282

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.2</td>

283

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.5</td>

284

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.1</td>

285

</tr>

286

<tr>

287

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HMMT Feb 26</td>

288

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.3</td>

289

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.2</td>

290

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">78.7</td>

291

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.0</td>

292

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.6</td>

293

</tr>

294

<tr>

295

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">IMOAnswerBench</td>

296

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.9</td>

297

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.5</td>

298

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.8</td>

299

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.3</td>

300

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">78.9</td>

301

</tr>

302

<tr>

303

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">AIME26 </td>

304

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.6</td>

305

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.2</td>

306

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">91.0</td>

307

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">88.3</td>

308

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.7</td>

309

</tr>

310

</tbody>

311

</table>

312

313

<p style="margin-top:12px;font-size:10px;opacity:0.7">

314

* 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/>

315

* 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/>

316

* SkillsBench: Evaluated via OpenCode on 78 tasks (self-contained subset, excluding API-dependent tasks); avg of 5 runs.<br/>

317

* NL2Repo: Others are evaluated via Claude Code (temp=1.0, top_p=0.95, max_turns=900).<br/>

318

* QwenClawBench: An internal real-user-distribution Claw agent benchmark (open-sourcing soon); temp=0.6, 256K ctx.<br/>

319

* 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/>

320

* TAU3-Bench: We use the official user model (gpt-5.2, low reasoning effort) + default BM25 retrieval.<br/>

321

* VITA-Bench: Avg subdomain scores; using claude-4-sonnet as judger, as the official judger (claude-3.7-sonnet) is no longer available.<br/>

322

* MCPMark: GitHub MCP v0.30.3; Playwright responses truncated at 32K tokens.<br/>

323

* MCP-Atlas: Public set score; gemini-2.5-pro judger.<br/>

324

* AIME 26: We use the full AIME 2026 (I & II), where the scores may differ from Qwen 3.5 notes.<br/>

325

</p>

326

327

</div>

328

329

330

### Vision Language

331

332

<div style="font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif;max-width:1000px;margin:0 auto;padding:16px 0">

333

<table style="width:100%;border-collapse:collapse;font-size:13px">

334

<thead><tr>

335

<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;">Claude-Sonnet-4.5</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;">Gemma4-26BA4B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.5-35B-A3B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.6-35B-A3B</th></tr></thead>

336

<tbody>

337

<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 and Puzzle</td></tr>

338

<tr>

339

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMMU</td>

340

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.3</td>

341

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.6</td>

342

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.4</td>

343

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">78.4</td>

344

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.4</td>

345

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.7</td>

346

</tr>

347

<tr>

348

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMMU-Pro</td>

349

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.0</td>

350

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">68.4</td>

351

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.9*</td>

352

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">73.8*</td>

353

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.1</td>

354

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.3</td>

355

</tr>

356

<tr>

357

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Mathvista(mini)</td>

358

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.8</td>

359

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.8</td>

360

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.3</td>

361

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.4</td>

362

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.2</td>

363

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.4</td>

364

</tr>

365

<tr>

366

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">ZEROBench_sub</td>

367

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">36.2</td>

368

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">26.3</td>

369

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">26.0</td>

370

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">26.3</td>

371

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">34.1</td>

372

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">34.4</td>

373

</tr>

374

<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>

375

<tr>

376

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">RealWorldQA</td>

377

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.7</td>

378

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.3</td>

379

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">72.3</td>

380

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">72.2</td>

381

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.1</td>

382

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.3</td>

383

</tr>

384

<tr>

385

<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>

386

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.6</td>

387

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">88.3</td>

388

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.9</td>

389

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.0</td>

390

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">91.5</td>

391

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.8</td>

392

</tr>

393

<tr>

394

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SimpleVQA</td>

395

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">56.0</td>

396

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">57.6</td>

397

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.9</td>

398

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.2</td>

399

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">58.3</td>

400

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">58.9</td>

401

</tr>

402

<tr>

403

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HallusionBench</td>

404

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.0</td>

405

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">59.9</td>

406

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.4</td>

407

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">66.1</td>

408

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.9</td>

409

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">69.8</td>

410

</tr>

411

<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)">Text Recognition and Document Understanding</td></tr>

412

<tr>

413

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">OmniDocBench1.5</td>

414

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">88.9</td>

415

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.8</td>

416

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.1</td>

417

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.4</td>

418

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.3</td>

419

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.9</td>

420

</tr>

421

<tr>

422

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">CharXiv(RQ)</td>

423

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.5</td>

424

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.2</td>

425

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.9</td>

426

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">69.0</td>

427

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.5</td>

428

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">78.0</td>

429

</tr>

430

<tr>

431

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">CC-OCR</td>

432

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.0</td>

433

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">68.1</td>

434

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.7</td>

435

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.5</td>

436

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.7</td>

437

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.9</td>

438

</tr>

439

<tr>

440

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">AI2D_TEST</td>

441

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.9</td>

442

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.0</td>

443

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.0</td>

444

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">88.3</td>

445

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.6</td>

446

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.7</td>

447

</tr>

448

<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>

449

<tr>

450

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">RefCOCO(avg)</td>

451

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.9</td>

452

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

453

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</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)">89.2</td>

456

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.0</td>

457

</tr>

458

<tr>

459

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">ODInW13</td>

460

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">41.1</td>

461

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

462

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

463

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

464

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">42.6</td>

465

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">50.8</td>

466

</tr>

467

<tr>

468

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">EmbSpatialBench</td>

469

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.5</td>

470

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">71.8</td>

471

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

472

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

473

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.1</td>

474

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.3</td>

475

</tr>

476

<tr>

477

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">RefSpatialBench</td>

478

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.7</td>

479

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

480

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

481

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

482

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">63.5</td>

483

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.3</td>

484

</tr>

485

<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>

486

<tr>

487

<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>

488

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.0</td>

489

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.1</td>

490

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</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)">86.6</td>

493

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.6</td>

494

</tr>

495

<tr>

496

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">VideoMME<sub><small>(w/o sub.)</sub></small></td>

497

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.8</td>

498

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.3</td>

499

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</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)">82.5</td>

502

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.5</td>

503

</tr>

504

<tr>

505

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">VideoMMMU</td>

506

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.3</td>

507

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.6</td>

508

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.6</td>

509

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.0</td>

510

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.4</td>

511

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.7</td>

512

</tr>

513

<tr>

514

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MLVU</td>

515

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.9</td>

516

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">72.8</td>

517

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

518

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

519

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.6</td>

520

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.2</td>

521

</tr>

522

<tr>

523

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MVBench</td>

524

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.6</td>

525

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

526

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

527

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

528

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.8</td>

529

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.6</td>

530

</tr>

531

<tr>

532

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">LVBench</td>

533

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">73.6</td>

534

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

535

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

536

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>

537

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">71.4</td>

538

<td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">71.4</td>

539

</tr>

540

</tbody>

541

</table>

542

<p style="margin-top:12px;font-size:10px;opacity:0.7">

543

* Empty cells (--) indicate scores not available or not applicable.

544

</p>

545

</div>

546

547

## Quickstart

548

549

For streamlined integration, we recommend using Qwen3.6 via APIs. Below is a guide to use Qwen3.6 via OpenAI-compatible API.

550

551

### Serving Qwen3.6

552

553

Qwen3.6 can be served via APIs with popular inference frameworks.

554

In the following, we show example commands to launch OpenAI-Compatible API servers for Qwen3.6 models.

555

556

> [!Important]

557

> Inference efficiency and throughput vary significantly across frameworks.

558

> We recommend using the latest framework versions to ensure optimal performance and compatibility.

559

> For production workloads or high-throughput scenarios, dedicated serving engines such as SGLang, KTransformers or vLLM are strongly recommended.

560

561

> [!Important]

562

> The model has a default context length of 262,144 tokens.

563

> If you encounter out-of-memory (OOM) errors, consider reducing the context window.

564

> 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.

565

566

#### SGLang

567

568

[SGLang](https://github.com/sgl-project/sglang) is a fast serving framework for large language models and vision language models.

569

`sglang>=0.5.10` is recommended for Qwen3.6, which can be installed using the following command in a fresh environment:

570

```shell

571

uv pip install sglang[all]

572

```

573

See [its documentation](https://docs.sglang.ai/get_started/install.html) for more details.

574

575

The following will create API endpoints at `http://localhost:8000/v1`:

576

577

- **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.

578

579

```shell

580

    python -m sglang.launch_server --model-path Qwen/Qwen3.6-35B-A3B-FP8 --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3

581

```

582

583

- **Tool Use**: To support tool use, you can use the following command.

584

585

```shell

586

    python -m sglang.launch_server --model-path Qwen/Qwen3.6-35B-A3B-FP8 --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3 --tool-call-parser qwen3_coder

587

```

588

589

- **Multi-Token Prediction (MTP)**: The following command is recommended for MTP:

590

591

```shell

592

    python -m sglang.launch_server --model-path Qwen/Qwen3.6-35B-A3B-FP8 --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

593

```

594

595

For detailed deployment guide, see the [SGLang Qwen3.5 Cookbook](https://lmsysorg.mintlify.app/cookbook/llm/Qwen/Qwen3.5).

596

597

#### vLLM

598

599

[vLLM](https://github.com/vllm-project/vllm) is a high-throughput and memory-efficient inference and serving engine for LLMs.

600

`vllm>=0.19.0` is recommended for Qwen3.6, which can be installed using the following command in a fresh environment:

601

```shell

602

uv pip install vllm --torch-backend=auto

603

```

604

See [its documentation](https://docs.vllm.ai/en/stable/getting_started/installation/index.html) for more details.

605

606

607

The following will create API endpoints at `http://localhost:8000/v1`:

608

609

- **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.

610

611

```shell

612

    vllm serve Qwen/Qwen3.6-35B-A3B-FP8 --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3

613

```

614

615

- **Tool Call**: To support tool use, you can use the following command.

616

617

```shell

618

    vllm serve Qwen/Qwen3.6-35B-A3B-FP8 --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder

619

```

620

621

- **Multi-Token Prediction (MTP)**: The following command is recommended for MTP:

622

623

```shell

624

    vllm serve Qwen/Qwen3.6-35B-A3B-FP8 --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --speculative-config '{"method":"qwen3_next_mtp","num_speculative_tokens":2}'

625

```

626

627

- **Text-Only**: The following command skips the vision encoder and multimodal profiling to free up memory for additional KV cache:

628

629

```shell

630

    vllm serve Qwen/Qwen3.6-35B-A3B-FP8 --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --language-model-only

631

```

632

633

For detailed deployment guide, see the [vLLM Qwen3.5 Recipe](https://docs.vllm.ai/projects/recipes/en/latest/Qwen/Qwen3.5.html).

634

635

#### KTransformers

636

637

[KTransformers](https://github.com/kvcache-ai/ktransformers) is a flexible framework for experiencing cutting-edge LLM inference optimizations with CPU-GPU heterogeneous computing.

638

For running Qwen3.6 with KTransformers, see the [KTransformers Deployment Guide](https://github.com/kvcache-ai/ktransformers/blob/main/doc/en/Qwen3.5.md).

639

640

#### Hugging Face Transformers

641

642

Hugging Face Transformers contains a _lightweight_ server which can be used for quick testing and moderate load deployment.

643

The latest `transformers` is required for Qwen3.6:

644

```shell

645

pip install "transformers[serving]"

646

```

647

See [its documentation](https://huggingface.co/docs/transformers/main/serving) for more details. Please also make sure torchvision and pillow are installed.

648

649

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:

650

```shell

651

transformers serve Qwen/Qwen3.6-35B-A3B-FP8 --port 8000 --continuous-batching

652

```

653

654

### Using Qwen3.6 via the Chat Completions API

655

656

The chat completions API is accessible via standard HTTP requests or OpenAI SDKs.

657

Here, we show examples using the OpenAI Python SDK.

658

659

Before starting, make sure it is installed and the API key and the API base URL is configured, e.g.:

660

```shell

661

pip install -U openai

662

663

# Set the following accordingly

664

export OPENAI_BASE_URL="http://localhost:8000/v1"

665

export OPENAI_API_KEY="EMPTY"

666

```

667

668

> [!Tip]

669

> We recommend using the following set of sampling parameters for generation

670

> - Thinking mode for general tasks: `temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0`

671

> - 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`

672

> - 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`

673

>

674

> Please note that the support for sampling parameters varies according to inference frameworks.

675

676

> [!Important]

677

> Qwen3.6 models operate in thinking mode by default, generating thinking content signified by `<think>\n...</think>\n\n` before producing the final responses.

678

> To disable thinking content and obtain direct response, refer to the examples [here](#instruct-or-non-thinking-mode).

679

680

681

#### Text-Only Input

682

683

```python

684

from openai import OpenAI

685

# Configured by environment variables

686

client = OpenAI()

687

688

messages = [

689

{"role": "user", "content": "Type \"I love Qwen3.6\" backwards"},

690

]

691

692

chat_response = client.chat.completions.create(

693

model="Qwen/Qwen3.6-35B-A3B-FP8",

694

messages=messages,

695

max_tokens=81920,

696

temperature=1.0,

697

top_p=0.95,

698

presence_penalty=1.5,

699

extra_body={

700

"top_k": 20,

701

},

702

)

703

print("Chat response:", chat_response)

704

```

705

706

707

#### Image Input

708

709

```python

710

from openai import OpenAI

711

# Configured by environment variables

712

client = OpenAI()

713

714

messages = [

715

{

716

"role": "user",

717

"content": [

718

{

719

"type": "image_url",

720

"image_url": {

721

"url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/CI_Demo/mathv-1327.jpg"

722

}

723

},

724

{

725

"type": "text",

726

                "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) $1+\\sqrt{2}$"

727

}

728

]

729

}

730

]

731

732

chat_response = client.chat.completions.create(

733

model="Qwen/Qwen3.6-35B-A3B-FP8",

734

messages=messages,

735

max_tokens=81920,

736

temperature=1.0,

737

top_p=0.95,

738

presence_penalty=1.5,

739

extra_body={

740

"top_k": 20,

741

},

742

)

743

print("Chat response:", chat_response)

744

```

745

746

#### Video Input

747

748

```python

749

from openai import OpenAI

750

# Configured by environment variables

751

client = OpenAI()

752

753

messages = [

754

{

755

"role": "user",

756

"content": [

757

{

758

"type": "video_url",

759

"video_url": {

760

"url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/video/N1cdUjctpG8.mp4"

761

}

762

},

763

{

764

"type": "text",

765

                "text": "How many porcelain jars were discovered in the niches located in the primary chamber of the tomb?"

766

}

767

]

768

}

769

]

770

771

# When vLLM is launched with `--media-io-kwargs '{"video": {"num_frames": -1}}'`,

772

# video frame sampling can be configured via `extra_body` (e.g., by setting `fps`).

773

# This feature is currently supported only in vLLM.

774

#

775

# By default, `fps=2` and `do_sample_frames=True`.

776

# With `do_sample_frames=True`, you can customize the `fps` value to set your desired video sampling rate.

777

chat_response = client.chat.completions.create(

778

model="Qwen/Qwen3.6-35B-A3B-FP8",

779

messages=messages,

780

max_tokens=81920,

781

temperature=1.0,

782

top_p=0.95,

783

presence_penalty=1.5,

784

extra_body={

785

"top_k": 20,

786

"mm_processor_kwargs": {"fps": 2, "do_sample_frames": True},

787

},

788

)

789

790

print("Chat response:", chat_response)

791

```

792

793

794

#### Instruct (or Non-Thinking) Mode

795

796

> [!Important]

797

> Qwen3.6 does not officially support the soft switch of Qwen3, i.e., `/think` and `/nothink`.

798

799

Qwen3.6 will think by default before response.

800

You can obtain direct response from the model without thinking by configuring the API parameters.

801

For example,

802

```python

803

from openai import OpenAI

804

# Configured by environment variables

805

client = OpenAI()

806

807

messages = [

808

{

809

"role": "user",

810

"content": [

811

{

812

"type": "image_url",

813

"image_url": {

814

"url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.6/demo/RealWorld/RealWorld-04.png"

815

}

816

},

817

{

818

"type": "text",

819

"text": "Where is this?"

820

}

821

]

822

}

823

]

824

825

chat_response = client.chat.completions.create(

826

model="Qwen/Qwen3.6-35B-A3B-FP8",

827

messages=messages,

828

max_tokens=32768,

829

temperature=0.7,

830

top_p=0.8,

831

presence_penalty=1.5,

832

extra_body={

833

"top_k": 20,

834

"chat_template_kwargs": {"enable_thinking": False},

835

},

836

)

837

print("Chat response:", chat_response)

838

```

839

840

> [!Note]

841

> 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}`.

842

843

#### Preserve Thinking

844

845

By default, only the thinking blocks generated in handling the latest user message is retained, resulting in a pattern commonly as interleaved thinking.

846

Qwen3.6 has been additionally trained to preserve and leverage thinking traces from historical messages.

847

You can enable this behavior by setting the `preserve_thinking` option:

848

849

```python

850

from openai import OpenAI

851

# Configured by environment variables

852

client = OpenAI()

853

854

messages = [...]

855

856

chat_response = client.chat.completions.create(

857

model="Qwen/Qwen3.6-35B-A3B-FP8",

858

messages=messages,

859

max_tokens=32768,

860

temperature=0.7,

861

top_p=0.8,

862

presence_penalty=1.5,

863

extra_body={

864

"top_k": 20,

865

"chat_template_kwargs": {"preserve_thinking": True},

866

},

867

)

868

print("Chat response:", chat_response)

869

```

870

871

> [!Note]

872

> 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}`.

873

874

875

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.

876

877

878

## Agentic Usage

879

880

Qwen3.6 excels in tool calling capabilities.

881

882

### Qwen-Agent

883

884

We recommend using [Qwen-Agent](https://github.com/QwenLM/Qwen-Agent) to quickly build Agent applications with Qwen3.6.

885

886

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.

887

```python

888

import os

889

from qwen_agent.agents import Assistant

890

891

# Define LLM

892

# Using Alibaba Cloud Model Studio

893

llm_cfg = {

894

# Use the OpenAI-compatible model service provided by DashScope:

895

'model': 'Qwen3.6-35B-A3B',

896

'model_type': 'qwenvl_oai',

897

'model_server': 'https://dashscope.aliyuncs.com/compatible-mode/v1',

898

'api_key': os.getenv('DASHSCOPE_API_KEY'),

899

900

'generate_cfg': {

901

'use_raw_api': True,

902

# When using Dash Scope OAI API, pass the parameter of whether to enable thinking mode in this way

903

'extra_body': {

904

'enable_thinking': True,

905

'preserve_thinking': True,

906

},

907

},

908

}

909

910

# Using OpenAI-compatible API endpoint.

911

# functionality of the deployment frameworks and let Qwen-Agent automate the related operations.

912

#

913

# llm_cfg = {

914

# # Use your own model service compatible with OpenAI API by vLLM/SGLang:

915

# 'model': 'Qwen/Qwen3.6-35B-A3B-FP8',

916

# 'model_type': 'qwenvl_oai',

917

# 'model_server': 'http://localhost:8000/v1', # api_base

918

# 'api_key': 'EMPTY',

919

#

920

# 'generate_cfg': {

921

# 'use_raw_api': True,

922

# # When using vLLM/SGLang OAI API, pass the parameter of whether to enable thinking mode in this way

923

# 'extra_body': {

924

# 'chat_template_kwargs': {'enable_thinking': True, 'preserve_thinking': True}

925

# },

926

# },

927

# }

928

929

# Define Tools

930

tools = [

931

{'mcpServers': { # You can specify the MCP configuration file

932

"filesystem": {

933

"command": "npx",

934

"args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/xxxx/Desktop"]

935

}

936

}

937

}

938

]

939

940

# Define Agent

941

bot = Assistant(llm=llm_cfg, function_list=tools)

942

943

# Streaming generation

944

messages = [{'role': 'user', 'content': 'Help me organize my desktop.'}]

945

for responses in bot.run(messages=messages):

946

pass

947

print(responses)

948

949

# Streaming generation

950

messages = [{'role': 'user', 'content': 'Develop a dog website and save it on the desktop'}]

951

for responses in bot.run(messages=messages):

952

pass

953

print(responses)

954

```

955

956

### Qwen Code

957

958

959

[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.

960

961

For more information, please refer to [Qwen Code](https://qwenlm.github.io/qwen-code-docs/).

962

963

## Processing Ultra-Long Texts

964

965

Qwen3.6 natively supports context lengths of up to 262,144 tokens.

966

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.

967

968

YaRN is currently supported by several inference frameworks, e.g., `transformers`, `vllm`, `ktransformers` and `sglang`.

969

In general, there are two approaches to enabling YaRN for supported frameworks:

970

971

- Modifying the model configuration file:

972

In the `config.json` file, change the `rope_parameters` fields in `text_config` to:

973

```json

974

{

975

"mrope_interleaved": true,

976

"mrope_section": [

977

11,

978

11,

979

10

980

],

981

"rope_type": "yarn",

982

"rope_theta": 10000000,

983

"partial_rotary_factor": 0.25,

984

"factor": 4.0,

985

"original_max_position_embeddings": 262144,

986

}

987

```

988

989

- Passing command line arguments:

990

991

For `vllm`, you can use

992

```shell

993

    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

994

```

995

996

For `sglang` and `ktransformers`, you can use

997

```shell

998

    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

999

```

1000

1001

> [!NOTE]

1002

> 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.**

1003

> We advise modifying the `rope_parameters` configuration only when processing long contexts is required.

1004

> 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.

1005

1006

## Best Practices

1007

1008

To achieve optimal performance, we recommend the following settings:

1009

1010

1. **Sampling Parameters**:

1011

- We suggest using the following sets of sampling parameters depending on the mode and task type:

1012

- **Thinking mode for general tasks**:

1013

`temperature=1.0`, `top_p=0.95`, `top_k=20`, `min_p=0.0`, `presence_penalty=1.5`, `repetition_penalty=1.0`

1014

- **Thinking mode for precise coding tasks (e.g., WebDev)**:

1015

`temperature=0.6`, `top_p=0.95`, `top_k=20`, `min_p=0.0`, `presence_penalty=0.0`, `repetition_penalty=1.0`

1016

- **Instruct (or non-thinking) mode**:

1017

`temperature=0.7`, `top_p=0.80`, `top_k=20`, `min_p=0.0`, `presence_penalty=1.5`, `repetition_penalty=1.0`

1018

   - 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.

1019

1020

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.

1021

1022

3. **Standardize Output Format**: We recommend using prompts to standardize model outputs when benchmarking.

1023

- **Math Problems**: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt.

1024

   - **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"`."

1025

1026

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,

1027

```json

1028

{"longest_edge": 469762048, "shortest_edge": 4096}

1029

```

1030

1031

    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).

1032

1033

1034

### Citation

1035

1036

If you find our work helpful, feel free to give us a cite.

1037

1038

```bibtex

1039

@misc{qwen36_35b_a3b,

1040

title = {{Qwen3.6-35B-A3B}: Agentic Coding Power, Now Open to All},

1041

url = {https://qwen.ai/blog?id=qwen3.6-35b-a3b},

1042

author = {{Qwen Team}},

1043

month = {April},

1044

year = {2026}

1045

}

1046

```

1047