Upload README_UPSTREAM.md with huggingface_hub
Browse files- README_UPSTREAM.md +627 -0
README_UPSTREAM.md
ADDED
|
@@ -0,0 +1,627 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- compressed-tensors
|
| 4 |
+
license: other
|
| 5 |
+
license_name: modified-mit
|
| 6 |
+
library_name: transformers
|
| 7 |
+
pipeline_tag: image-text-to-text
|
| 8 |
+
---
|
| 9 |
+
<div align="center">
|
| 10 |
+
<picture>
|
| 11 |
+
<img src="figures/kimi-logo.png" width="30%" alt="Kimi K2.6">
|
| 12 |
+
</picture>
|
| 13 |
+
</div>
|
| 14 |
+
<hr>
|
| 15 |
+
<div align="center" style="line-height:1">
|
| 16 |
+
<a href="https://www.kimi.com" target="_blank"><img alt="Chat" src="https://img.shields.io/badge/🤖%20Chat-Kimi%20K2.6-ff6b6b?color=1783ff&logoColor=white"/></a>
|
| 17 |
+
<a href="https://www.moonshot.ai" target="_blank"><img alt="Homepage" src="https://img.shields.io/badge/Homepage-Moonshot%20AI-white?logo=Kimi&logoColor=white"/></a>
|
| 18 |
+
</div>
|
| 19 |
+
|
| 20 |
+
<div align="center" style="line-height: 1;">
|
| 21 |
+
<a href="https://huggingface.co/moonshotai" target="_blank"><img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Moonshot%20AI-ffc107?color=ffc107&logoColor=white"/></a>
|
| 22 |
+
<a href="https://twitter.com/kimi_moonshot" target="_blank"><img alt="Twitter Follow" src="https://img.shields.io/badge/Twitter-Kimi.ai-white?logo=x&logoColor=white"/></a>
|
| 23 |
+
<a href="https://discord.gg/TYU2fdJykW" target="_blank"><img alt="Discord" src="https://img.shields.io/badge/Discord-Kimi.ai-white?logo=discord&logoColor=white"/></a>
|
| 24 |
+
<a href="https://modelscope.cn/organization/moonshotai" target="_blank"><img alt="ModelScope" src="https://img.shields.io/badge/ModelScope-Moonshot%20AI-white?labelColor=rgb(99%2C%2074%2C%20255)"/></a>
|
| 25 |
+
</div>
|
| 26 |
+
<div align="center" style="line-height: 1;">
|
| 27 |
+
<a href="https://huggingface.co/moonshotai/Kimi-K2.6/blob/main/LICENSE"><img alt="License" src="https://img.shields.io/badge/License-Modified_MIT-f5de53?&color=f5de53"/></a>
|
| 28 |
+
</div>
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
<p align="center">
|
| 32 |
+
🤗 <a href="https://huggingface.co/spaces/akhaliq/Kimi-K2.6" target="_blank">huggingchat</a>
|
| 33 |
+
|
|
| 34 |
+
📰 <a href="https://www.kimi.com/blog/kimi-k2-6.html">Tech Blog</a>
|
| 35 |
+
</p>
|
| 36 |
+
|
| 37 |
+
|
| 38 |
+
## 1. Model Introduction
|
| 39 |
+
|
| 40 |
+
Kimi K2.6 is an open-source, native multimodal agentic model that advances practical capabilities in long-horizon coding, coding-driven design, proactive autonomous execution, and swarm-based task orchestration.
|
| 41 |
+
|
| 42 |
+
### Key Features
|
| 43 |
+
- **Long-Horizon Coding**: K2.6 achieves significant improvements on complex, end-to-end coding tasks, generalizing robustly across programming languages (Rust, Go, Python) and domains spanning front-end, DevOps, and performance optimization.
|
| 44 |
+
- **Coding-Driven Design**: K2.6 is capable of transforming simple prompts and visual inputs into production-ready interfaces and lightweight full-stack workflows, generating structured layouts, interactive elements, and rich animations with deliberate aesthetic precision.
|
| 45 |
+
- **Elevated Agent Swarm**: Scaling horizontally to 300 sub-agents executing 4,000 coordinated steps, K2.6 can dynamically decompose tasks into parallel, domain-specialized subtasks, delivering end-to-end outputs from documents to websites to spreadsheets in a single autonomous run.
|
| 46 |
+
- **Proactive & Open Orchestration**: For autonomous tasks, K2.6 demonstrates strong performance in powering persistent, 24/7 background agents that proactively manage schedules, execute code, and orchestrate cross-platform operations without human oversight.
|
| 47 |
+
|
| 48 |
+
## 2. Model Summary
|
| 49 |
+
|
| 50 |
+
<div align="center">
|
| 51 |
+
|
| 52 |
+
|
| 53 |
+
| | |
|
| 54 |
+
|:---:|:---:|
|
| 55 |
+
| **Architecture** | Mixture-of-Experts (MoE) |
|
| 56 |
+
| **Total Parameters** | 1T |
|
| 57 |
+
| **Activated Parameters** | 32B |
|
| 58 |
+
| **Number of Layers** (Dense layer included) | 61 |
|
| 59 |
+
| **Number of Dense Layers** | 1 |
|
| 60 |
+
| **Attention Hidden Dimension** | 7168 |
|
| 61 |
+
| **MoE Hidden Dimension** (per Expert) | 2048 |
|
| 62 |
+
| **Number of Attention Heads** | 64 |
|
| 63 |
+
| **Number of Experts** | 384 |
|
| 64 |
+
| **Selected Experts per Token** | 8 |
|
| 65 |
+
| **Number of Shared Experts** | 1 |
|
| 66 |
+
| **Vocabulary Size** | 160K |
|
| 67 |
+
| **Context Length** | 256K |
|
| 68 |
+
| **Attention Mechanism** | MLA |
|
| 69 |
+
| **Activation Function** | SwiGLU |
|
| 70 |
+
| **Vision Encoder** | MoonViT |
|
| 71 |
+
| **Parameters of Vision Encoder** | 400M |
|
| 72 |
+
</div>
|
| 73 |
+
|
| 74 |
+
## 3. Evaluation Results
|
| 75 |
+
|
| 76 |
+
<div align="center">
|
| 77 |
+
<table>
|
| 78 |
+
<thead>
|
| 79 |
+
<tr>
|
| 80 |
+
<th align="center">Benchmark</th>
|
| 81 |
+
<th align="center"><sup>Kimi K2.6</sup></th>
|
| 82 |
+
<th align="center"><sup>GPT-5.4 <br><sup>(xhigh)</sup></sup></th>
|
| 83 |
+
<th align="center"><sup>Claude Opus 4.6 <br><sup>(max effort)</sup></sup></th>
|
| 84 |
+
<th align="center"><sup>Gemini 3.1 Pro<br><sup>(thinking high)</sup></sup></th>
|
| 85 |
+
<th align="center"><sup>Kimi K2.5</sup></th>
|
| 86 |
+
</tr>
|
| 87 |
+
</thead>
|
| 88 |
+
<tbody>
|
| 89 |
+
<tr>
|
| 90 |
+
<td align="center" colspan=6><strong>Agentic</strong></td>
|
| 91 |
+
</tr>
|
| 92 |
+
<tr>
|
| 93 |
+
<td align="center" style="vertical-align: middle">HLE-Full<br>(w/ tools)</td>
|
| 94 |
+
<td align="center" style="vertical-align: middle">54.0</td>
|
| 95 |
+
<td align="center" style="vertical-align: middle">52.1</td>
|
| 96 |
+
<td align="center" style="vertical-align: middle">53.0</td>
|
| 97 |
+
<td align="center" style="vertical-align: middle">51.4</td>
|
| 98 |
+
<td align="center" style="vertical-align: middle">50.2</td>
|
| 99 |
+
</tr>
|
| 100 |
+
<tr>
|
| 101 |
+
<td align="center" style="vertical-align: middle">BrowseComp</td>
|
| 102 |
+
<td align="center" style="vertical-align: middle">83.2</td>
|
| 103 |
+
<td align="center" style="vertical-align: middle" rowspan="2">82.7</td>
|
| 104 |
+
<td align="center" style="vertical-align: middle" rowspan="2">83.7</td>
|
| 105 |
+
<td align="center" style="vertical-align: middle" rowspan="2">85.9</td>
|
| 106 |
+
<td align="center" style="vertical-align: middle">74.9</td>
|
| 107 |
+
</tr>
|
| 108 |
+
<tr>
|
| 109 |
+
<td align="center" style="vertical-align: middle">BrowseComp<br>(Agent Swarm)</td>
|
| 110 |
+
<td align="center" style="vertical-align: middle">86.3</td>
|
| 111 |
+
<td align="center" style="vertical-align: middle">78.4</td>
|
| 112 |
+
</tr>
|
| 113 |
+
<tr>
|
| 114 |
+
<td align="center" style="vertical-align: middle">DeepSearchQA<br>(f1-score)</td>
|
| 115 |
+
<td align="center" style="vertical-align: middle">92.5</td>
|
| 116 |
+
<td align="center" style="vertical-align: middle">78.6</td>
|
| 117 |
+
<td align="center" style="vertical-align: middle">91.3</td>
|
| 118 |
+
<td align="center" style="vertical-align: middle">81.9</td>
|
| 119 |
+
<td align="center" style="vertical-align: middle">89.0</td>
|
| 120 |
+
</tr>
|
| 121 |
+
<tr>
|
| 122 |
+
<td align="center" style="vertical-align: middle">DeepSearchQA<br>(accuracy)</td>
|
| 123 |
+
<td align="center" style="vertical-align: middle">83.0</td>
|
| 124 |
+
<td align="center" style="vertical-align: middle">63.7</td>
|
| 125 |
+
<td align="center" style="vertical-align: middle">80.6</td>
|
| 126 |
+
<td align="center" style="vertical-align: middle">60.2</td>
|
| 127 |
+
<td align="center" style="vertical-align: middle">77.1</td>
|
| 128 |
+
</tr>
|
| 129 |
+
<tr>
|
| 130 |
+
<td align="center" style="vertical-align: middle">WideSearch<br> (item-f1)</td>
|
| 131 |
+
<td align="center" style="vertical-align: middle">80.8</td>
|
| 132 |
+
<td align="center" style="vertical-align: middle">-</td>
|
| 133 |
+
<td align="center" style="vertical-align: middle">-</td>
|
| 134 |
+
<td align="center" style="vertical-align: middle">-</td>
|
| 135 |
+
<td align="center" style="vertical-align: middle">72.7</td>
|
| 136 |
+
</tr>
|
| 137 |
+
<tr>
|
| 138 |
+
<td align="center" style="vertical-align: middle">Toolathlon</td>
|
| 139 |
+
<td align="center" style="vertical-align: middle">50.0</td>
|
| 140 |
+
<td align="center" style="vertical-align: middle">54.6</td>
|
| 141 |
+
<td align="center" style="vertical-align: middle">47.2</td>
|
| 142 |
+
<td align="center" style="vertical-align: middle">48.8</td>
|
| 143 |
+
<td align="center" style="vertical-align: middle">27.8</td>
|
| 144 |
+
</tr>
|
| 145 |
+
<tr>
|
| 146 |
+
<td align="center" style="vertical-align: middle">MCPMark</td>
|
| 147 |
+
<td align="center" style="vertical-align: middle">55.9</td>
|
| 148 |
+
<td align="center" style="vertical-align: middle">62.5*</td>
|
| 149 |
+
<td align="center" style="vertical-align: middle">56.7*</td>
|
| 150 |
+
<td align="center" style="vertical-align: middle">55.9*</td>
|
| 151 |
+
<td align="center" style="vertical-align: middle">29.5</td>
|
| 152 |
+
</tr>
|
| 153 |
+
<tr>
|
| 154 |
+
<td align="center" style="vertical-align: middle">Claw Eval (pass^3)</td>
|
| 155 |
+
<td align="center" style="vertical-align: middle">62.3</td>
|
| 156 |
+
<td align="center" style="vertical-align: middle">60.3</td>
|
| 157 |
+
<td align="center" style="vertical-align: middle">70.4</td>
|
| 158 |
+
<td align="center" style="vertical-align: middle">57.8</td>
|
| 159 |
+
<td align="center" style="vertical-align: middle">52.3</td>
|
| 160 |
+
</tr>
|
| 161 |
+
<tr>
|
| 162 |
+
<td align="center" style="vertical-align: middle">Claw Eval (pass@3)</td>
|
| 163 |
+
<td align="center" style="vertical-align: middle">80.9</td>
|
| 164 |
+
<td align="center" style="vertical-align: middle">78.4</td>
|
| 165 |
+
<td align="center" style="vertical-align: middle">82.4</td>
|
| 166 |
+
<td align="center" style="vertical-align: middle">82.9</td>
|
| 167 |
+
<td align="center" style="vertical-align: middle">75.4</td>
|
| 168 |
+
</tr>
|
| 169 |
+
<tr>
|
| 170 |
+
<td align="center" style="vertical-align: middle">APEX-Agents</td>
|
| 171 |
+
<td align="center" style="vertical-align: middle">27.9</td>
|
| 172 |
+
<td align="center" style="vertical-align: middle">33.3</td>
|
| 173 |
+
<td align="center" style="vertical-align: middle">33.0</td>
|
| 174 |
+
<td align="center" style="vertical-align: middle">32.0</td>
|
| 175 |
+
<td align="center" style="vertical-align: middle">11.5</td>
|
| 176 |
+
</tr>
|
| 177 |
+
<tr>
|
| 178 |
+
<td align="center" style="vertical-align: middle">OSWorld-Verified</td>
|
| 179 |
+
<td align="center" style="vertical-align: middle">73.1</td>
|
| 180 |
+
<td align="center" style="vertical-align: middle">75.0</td>
|
| 181 |
+
<td align="center" style="vertical-align: middle">72.7</td>
|
| 182 |
+
<td align="center" style="vertical-align: middle">-</td>
|
| 183 |
+
<td align="center" style="vertical-align: middle">63.3</td>
|
| 184 |
+
</tr>
|
| 185 |
+
<tr>
|
| 186 |
+
<td align="center" colspan=6><strong>Coding</strong></td>
|
| 187 |
+
</tr>
|
| 188 |
+
<tr>
|
| 189 |
+
<td align="center" style="vertical-align: middle">Terminal-Bench 2.0<br>(Terminus-2)</td>
|
| 190 |
+
<td align="center" style="vertical-align: middle">66.7</td>
|
| 191 |
+
<td align="center" style="vertical-align: middle">65.4*</td>
|
| 192 |
+
<td align="center" style="vertical-align: middle">65.4</td>
|
| 193 |
+
<td align="center" style="vertical-align: middle">68.5</td>
|
| 194 |
+
<td align="center" style="vertical-align: middle">50.8</td>
|
| 195 |
+
</tr>
|
| 196 |
+
<tr>
|
| 197 |
+
<td align="center" style="vertical-align: middle">SWE-Bench Pro</td>
|
| 198 |
+
<td align="center" style="vertical-align: middle">58.6</td>
|
| 199 |
+
<td align="center" style="vertical-align: middle">57.7</td>
|
| 200 |
+
<td align="center" style="vertical-align: middle">53.4</td>
|
| 201 |
+
<td align="center" style="vertical-align: middle">54.2</td>
|
| 202 |
+
<td align="center" style="vertical-align: middle">50.7</td>
|
| 203 |
+
</tr>
|
| 204 |
+
<tr>
|
| 205 |
+
<td align="center" style="vertical-align: middle">SWE-Bench Multilingual</td>
|
| 206 |
+
<td align="center" style="vertical-align: middle">76.7</td>
|
| 207 |
+
<td align="center" style="vertical-align: middle">-</td>
|
| 208 |
+
<td align="center" style="vertical-align: middle">77.8</td>
|
| 209 |
+
<td align="center" style="vertical-align: middle">76.9*</td>
|
| 210 |
+
<td align="center" style="vertical-align: middle">73.0</td>
|
| 211 |
+
</tr>
|
| 212 |
+
<tr>
|
| 213 |
+
<td align="center" style="vertical-align: middle">SWE-Bench Verified</td>
|
| 214 |
+
<td align="center" style="vertical-align: middle">80.2</td>
|
| 215 |
+
<td align="center" style="vertical-align: middle">-</td>
|
| 216 |
+
<td align="center" style="vertical-align: middle">80.8</td>
|
| 217 |
+
<td align="center" style="vertical-align: middle">80.6</td>
|
| 218 |
+
<td align="center" style="vertical-align: middle">76.8</td>
|
| 219 |
+
</tr>
|
| 220 |
+
<tr>
|
| 221 |
+
<td align="center" style="vertical-align: middle">SciCode</td>
|
| 222 |
+
<td align="center" style="vertical-align: middle">52.2</td>
|
| 223 |
+
<td align="center" style="vertical-align: middle">56.6</td>
|
| 224 |
+
<td align="center" style="vertical-align: middle">51.9</td>
|
| 225 |
+
<td align="center" style="vertical-align: middle">58.9</td>
|
| 226 |
+
<td align="center" style="vertical-align: middle">48.7</td>
|
| 227 |
+
</tr>
|
| 228 |
+
<tr>
|
| 229 |
+
<td align="center" style="vertical-align: middle">OJBench (python)</td>
|
| 230 |
+
<td align="center" style="vertical-align: middle">60.6</td>
|
| 231 |
+
<td align="center" style="vertical-align: middle">-</td>
|
| 232 |
+
<td align="center" style="vertical-align: middle">60.3</td>
|
| 233 |
+
<td align="center" style="vertical-align: middle">70.7</td>
|
| 234 |
+
<td align="center" style="vertical-align: middle">54.7</td>
|
| 235 |
+
</tr>
|
| 236 |
+
<tr>
|
| 237 |
+
<td align="center" style="vertical-align: middle">LiveCodeBench (v6)</td>
|
| 238 |
+
<td align="center" style="vertical-align: middle">89.6</td>
|
| 239 |
+
<td align="center" style="vertical-align: middle">-</td>
|
| 240 |
+
<td align="center" style="vertical-align: middle">88.8</td>
|
| 241 |
+
<td align="center" style="vertical-align: middle">91.7</td>
|
| 242 |
+
<td align="center" style="vertical-align: middle">85.0</td>
|
| 243 |
+
</tr>
|
| 244 |
+
<tr>
|
| 245 |
+
<td align="center" colspan=6><strong>Reasoning & Knowledge</strong></td>
|
| 246 |
+
</tr>
|
| 247 |
+
<tr>
|
| 248 |
+
<td align="center" style="vertical-align: middle">HLE-Full</td>
|
| 249 |
+
<td align="center" style="vertical-align: middle">34.7</td>
|
| 250 |
+
<td align="center" style="vertical-align: middle">39.8</td>
|
| 251 |
+
<td align="center" style="vertical-align: middle">40.0</td>
|
| 252 |
+
<td align="center" style="vertical-align: middle">44.4</td>
|
| 253 |
+
<td align="center" style="vertical-align: middle">30.1</td>
|
| 254 |
+
</tr>
|
| 255 |
+
<tr>
|
| 256 |
+
<td align="center" style="vertical-align: middle">AIME 2026</td>
|
| 257 |
+
<td align="center" style="vertical-align: middle">96.4</td>
|
| 258 |
+
<td align="center" style="vertical-align: middle">99.2</td>
|
| 259 |
+
<td align="center" style="vertical-align: middle">96.7</td>
|
| 260 |
+
<td align="center" style="vertical-align: middle">98.3</td>
|
| 261 |
+
<td align="center" style="vertical-align: middle">95.8</td>
|
| 262 |
+
</tr>
|
| 263 |
+
<tr>
|
| 264 |
+
<td align="center" style="vertical-align: middle">HMMT 2026 (Feb)</td>
|
| 265 |
+
<td align="center" style="vertical-align: middle">92.7</td>
|
| 266 |
+
<td align="center" style="vertical-align: middle">97.7</td>
|
| 267 |
+
<td align="center" style="vertical-align: middle">96.2</td>
|
| 268 |
+
<td align="center" style="vertical-align: middle">94.7</td>
|
| 269 |
+
<td align="center" style="vertical-align: middle">87.1</td>
|
| 270 |
+
</tr>
|
| 271 |
+
<tr>
|
| 272 |
+
<td align="center" style="vertical-align: middle">IMO-AnswerBench</td>
|
| 273 |
+
<td align="center" style="vertical-align: middle">86.0</td>
|
| 274 |
+
<td align="center" style="vertical-align: middle">91.4</td>
|
| 275 |
+
<td align="center" style="vertical-align: middle">75.3</td>
|
| 276 |
+
<td align="center" style="vertical-align: middle">91.0*</td>
|
| 277 |
+
<td align="center" style="vertical-align: middle">81.8</td>
|
| 278 |
+
</tr>
|
| 279 |
+
<tr>
|
| 280 |
+
<td align="center" style="vertical-align: middle">GPQA-Diamond</td>
|
| 281 |
+
<td align="center" style="vertical-align: middle">90.5</td>
|
| 282 |
+
<td align="center" style="vertical-align: middle">92.8</td>
|
| 283 |
+
<td align="center" style="vertical-align: middle">91.3</td>
|
| 284 |
+
<td align="center" style="vertical-align: middle">94.3</td>
|
| 285 |
+
<td align="center" style="vertical-align: middle">87.6</td>
|
| 286 |
+
</tr>
|
| 287 |
+
<tr>
|
| 288 |
+
<td align="center" colspan=6><strong>Vision</strong></td>
|
| 289 |
+
</tr>
|
| 290 |
+
<tr>
|
| 291 |
+
<td align="center" style="vertical-align: middle">MMMU-Pro</td>
|
| 292 |
+
<td align="center" style="vertical-align: middle">79.4</td>
|
| 293 |
+
<td align="center" style="vertical-align: middle">81.2</td>
|
| 294 |
+
<td align="center" style="vertical-align: middle">73.9</td>
|
| 295 |
+
<td align="center" style="vertical-align: middle">83.0*</td>
|
| 296 |
+
<td align="center" style="vertical-align: middle">78.5</td>
|
| 297 |
+
</tr>
|
| 298 |
+
<tr>
|
| 299 |
+
<td align="center" style="vertical-align: middle">MMMU-Pro (w/ python)</td>
|
| 300 |
+
<td align="center" style="vertical-align: middle">80.1</td>
|
| 301 |
+
<td align="center" style="vertical-align: middle">82.1</td>
|
| 302 |
+
<td align="center" style="vertical-align: middle">77.3</td>
|
| 303 |
+
<td align="center" style="vertical-align: middle">85.3*</td>
|
| 304 |
+
<td align="center" style="vertical-align: middle">77.7</td>
|
| 305 |
+
</tr>
|
| 306 |
+
<tr>
|
| 307 |
+
<td align="center" style="vertical-align: middle">CharXiv (RQ)</td>
|
| 308 |
+
<td align="center" style="vertical-align: middle">80.4</td>
|
| 309 |
+
<td align="center" style="vertical-align: middle">82.8*</td>
|
| 310 |
+
<td align="center" style="vertical-align: middle">69.1</td>
|
| 311 |
+
<td align="center" style="vertical-align: middle">80.2*</td>
|
| 312 |
+
<td align="center" style="vertical-align: middle">77.5</td>
|
| 313 |
+
</tr>
|
| 314 |
+
<tr>
|
| 315 |
+
<td align="center" style="vertical-align: middle">CharXiv (RQ) (w/ python)</td>
|
| 316 |
+
<td align="center" style="vertical-align: middle">86.7</td>
|
| 317 |
+
<td align="center" style="vertical-align: middle">90.0*</td>
|
| 318 |
+
<td align="center" style="vertical-align: middle">84.7</td>
|
| 319 |
+
<td align="center" style="vertical-align: middle">89.9*</td>
|
| 320 |
+
<td align="center" style="vertical-align: middle">78.7</td>
|
| 321 |
+
</tr>
|
| 322 |
+
<tr>
|
| 323 |
+
<td align="center" style="vertical-align: middle">MathVision</td>
|
| 324 |
+
<td align="center" style="vertical-align: middle">87.4</td>
|
| 325 |
+
<td align="center" style="vertical-align: middle">92.0*</td>
|
| 326 |
+
<td align="center" style="vertical-align: middle">71.2*</td>
|
| 327 |
+
<td align="center" style="vertical-align: middle">89.8*</td>
|
| 328 |
+
<td align="center" style="vertical-align: middle">84.2</td>
|
| 329 |
+
</tr>
|
| 330 |
+
<tr>
|
| 331 |
+
<td align="center" style="vertical-align: middle">MathVision (w/ python)</td>
|
| 332 |
+
<td align="center" style="vertical-align: middle">93.2</td>
|
| 333 |
+
<td align="center" style="vertical-align: middle">96.1*</td>
|
| 334 |
+
<td align="center" style="vertical-align: middle">84.6*</td>
|
| 335 |
+
<td align="center" style="vertical-align: middle">95.7*</td>
|
| 336 |
+
<td align="center" style="vertical-align: middle">85.0</td>
|
| 337 |
+
</tr>
|
| 338 |
+
<tr>
|
| 339 |
+
<td align="center" style="vertical-align: middle">BabyVision</td>
|
| 340 |
+
<td align="center" style="vertical-align: middle">39.8</td>
|
| 341 |
+
<td align="center" style="vertical-align: middle">49.7</td>
|
| 342 |
+
<td align="center" style="vertical-align: middle">14.8</td>
|
| 343 |
+
<td align="center" style="vertical-align: middle">51.6</td>
|
| 344 |
+
<td align="center" style="vertical-align: middle">36.5</td>
|
| 345 |
+
</tr>
|
| 346 |
+
<tr>
|
| 347 |
+
<td align="center" style="vertical-align: middle">BabyVision (w/ python)</td>
|
| 348 |
+
<td align="center" style="vertical-align: middle">68.5</td>
|
| 349 |
+
<td align="center" style="vertical-align: middle">80.2*</td>
|
| 350 |
+
<td align="center" style="vertical-align: middle">38.4*</td>
|
| 351 |
+
<td align="center" style="vertical-align: middle">68.3*</td>
|
| 352 |
+
<td align="center" style="vertical-align: middle">40.5</td>
|
| 353 |
+
</tr>
|
| 354 |
+
<tr>
|
| 355 |
+
<td align="center" style="vertical-align: middle">V* (w/ python)</td>
|
| 356 |
+
<td align="center" style="vertical-align: middle">96.9</td>
|
| 357 |
+
<td align="center" style="vertical-align: middle">98.4*</td>
|
| 358 |
+
<td align="center" style="vertical-align: middle">86.4*</td>
|
| 359 |
+
<td align="center" style="vertical-align: middle">96.9*</td>
|
| 360 |
+
<td align="center" style="vertical-align: middle">86.9</td>
|
| 361 |
+
</tr>
|
| 362 |
+
</tbody>
|
| 363 |
+
</table>
|
| 364 |
+
</div>
|
| 365 |
+
|
| 366 |
+
<details>
|
| 367 |
+
<summary><b>Footnotes</b></summary>
|
| 368 |
+
|
| 369 |
+
1. **General Testing Details**
|
| 370 |
+
- We report results for Kimi K2.6 and Kimi K2.5 with thinking mode enabled, Claude Opus 4.6 with max effort, GPT-5.4 with xhigh reasoning effort, and Gemini 3.1 Pro with a high thinking level.
|
| 371 |
+
- Unless otherwise specified, all Kimi K2.6 experiments were conducted with temperature = 1.0, top-p = 1.0, and a context length of 262,144 tokens.
|
| 372 |
+
- Benchmarks without publicly available scores were re-evaluated under the same conditions used for Kimi K2.6 and are marked with an asterisk (`*`). Except where noted with an asterisk, all other results are cited from official reports.
|
| 373 |
+
2. **Reasoning Benchmarks**
|
| 374 |
+
- IMO-AnswerBench scores for GPT-5.4 and Claude 4.6 were obtained from [z.ai/blog/glm-5.1](https://z.ai/blog/glm-5.1).
|
| 375 |
+
- Humanity's Last Exam (HLE) and other reasoning tasks were evaluated with a maximum generation length of 98,304 tokens. By default, we report results on the HLE full set. For the text-only subset, Kimi K2.6 achieves 36.4% accuracy without tools and 55.5% with tools.
|
| 376 |
+
3. **Tool-Augmented / Agentic Tasks**
|
| 377 |
+
- Kimi K2.6 was equipped with search, code-interpreter, and web-browsing tools for HLE with tools, BrowseComp, DeepSearchQA, and WideSearch.
|
| 378 |
+
- For HLE-Full with tools, the maximum generation length is 262,144 tokens with a per-step limit of 49,152 tokens. We employ a simple context management strategy: once the context window exceeds the threshold, only the most recent round of tool-related messages is retained.
|
| 379 |
+
- For BrowseComp, we report scores obtained with context management using the same discard-all strategy as Kimi K2.5 and DeepSeek-V3.2.
|
| 380 |
+
- For DeepSearchQA, no context management was applied to Kimi K2.6 tests, and tasks exceeding the supported context length were directly counted as failed. Scores for Claude Opus 4.6, GPT-5.4, and Gemini 3.1 Pro on DeepSearchQA are cited from the [Claude Opus 4.7 System Card](https://cdn.sanity.io/files/4zrzovbb/website/037f06850df7fbe871e206dad004c3db5fd50340.pdf).
|
| 381 |
+
- For WideSearch, we report results under the "hide tool result" context management setting. Once the context window exceeds the threshold, only the most recent round of tool-related messages is retained.
|
| 382 |
+
- The test system prompts are identical to those used in the [Kimi K2.5 technical report](https://arxiv.org/pdf/2602.02276).
|
| 383 |
+
- Claw Eval was conducted using version 1.1 with max-tokens-per-step = 16384.
|
| 384 |
+
- For APEX-Agents, we evaluate 452 tasks from the public 480-task release, as done by [Artificial Analysis](https://artificialanalysis.ai/evaluations/apex-agents-aa)(excluding Investment Banking Worlds 244 and 246, which have external runtime dependencies)
|
| 385 |
+
4. **Coding Tasks**
|
| 386 |
+
- Terminal-Bench 2.0 scores were obtained with the default agent framework (Terminus-2) and the provided JSON parser, operating in preserve thinking mode.
|
| 387 |
+
- For the SWE-Bench series of evaluations (including Verified, Multilingual, and Pro), we used an in-house evaluation framework adapted from SWE-agent. This framework includes a minimal set of tools—bash tool, createfile tool, insert tool, view tool, strreplace tool, and submit tool.
|
| 388 |
+
- All reported scores for coding tasks are averaged over 10 independent runs.
|
| 389 |
+
5. **Vision Benchmarks**
|
| 390 |
+
- Max-tokens = 98,304, averaged over three runs (avg@3).
|
| 391 |
+
- Settings with Python tool use max-tokens-per-step = 65,536 and max-steps = 50 for multi-step reasoning.
|
| 392 |
+
- MMMU-Pro follows the official protocol, preserving input order and prepending images.
|
| 393 |
+
|
| 394 |
+
</details>
|
| 395 |
+
|
| 396 |
+
|
| 397 |
+
## 4. Native INT4 Quantization
|
| 398 |
+
Kimi-K2.6 adopts the same native int4 quantization method as [Kimi-K2-Thinking](https://huggingface.co/moonshotai/Kimi-K2-Thinking#4-native-int4-quantization).
|
| 399 |
+
|
| 400 |
+
## 5. Deployment
|
| 401 |
+
|
| 402 |
+
> [!Note]
|
| 403 |
+
> You can access Kimi-K2.6's API on https://platform.moonshot.ai and we provide OpenAI/Anthropic-compatible API for you. To verify the deployment is correct, we also provide the [Kimi Vendor Verifier](https://kimi.com/blog/kimi-vendor-verifier.html).
|
| 404 |
+
Currently, Kimi-K2.6 is recommended to run on the following inference engines:
|
| 405 |
+
* vLLM
|
| 406 |
+
* SGLang
|
| 407 |
+
* KTransformers
|
| 408 |
+
|
| 409 |
+
Kimi-K2.6 has the same architecture as Kimi-K2.5, and the deployment method can be directly reused.
|
| 410 |
+
|
| 411 |
+
The version requirement for `transformers` is `>=4.57.1, <5.0.0`.
|
| 412 |
+
|
| 413 |
+
Deployment examples can be found in the [Model Deployment Guide](docs/deploy_guidance.md).
|
| 414 |
+
|
| 415 |
+
|
| 416 |
+
---
|
| 417 |
+
## 6. Model Usage
|
| 418 |
+
|
| 419 |
+
The usage demos below demonstrate how to call our official API.
|
| 420 |
+
|
| 421 |
+
For third-party APIs deployed with vLLM or SGLang, please note that:
|
| 422 |
+
> [!Note]
|
| 423 |
+
> - Chat with video content is an experimental feature and is only supported in our official API for now.
|
| 424 |
+
>
|
| 425 |
+
> - The recommended `temperature` will be `1.0` for Thinking mode and `0.6` for Instant mode.
|
| 426 |
+
>
|
| 427 |
+
> - The recommended `top_p` is `0.95`.
|
| 428 |
+
>
|
| 429 |
+
> - To use instant mode, you need to pass `{'chat_template_kwargs': {"thinking": False}}` in `extra_body`.
|
| 430 |
+
|
| 431 |
+
### Chat Completion
|
| 432 |
+
|
| 433 |
+
This is a simple chat completion script which shows how to call K2.6 API in Thinking and Instant modes.
|
| 434 |
+
|
| 435 |
+
```python
|
| 436 |
+
import openai
|
| 437 |
+
import base64
|
| 438 |
+
import requests
|
| 439 |
+
def simple_chat(client: openai.OpenAI, model_name: str):
|
| 440 |
+
messages = [
|
| 441 |
+
{'role': 'system', 'content': 'You are Kimi, an AI assistant created by Moonshot AI.'},
|
| 442 |
+
{
|
| 443 |
+
'role': 'user',
|
| 444 |
+
'content': [
|
| 445 |
+
{'type': 'text', 'text': 'which one is bigger, 9.11 or 9.9? think carefully.'}
|
| 446 |
+
],
|
| 447 |
+
},
|
| 448 |
+
]
|
| 449 |
+
response = client.chat.completions.create(
|
| 450 |
+
model=model_name, messages=messages, stream=False, max_tokens=4096
|
| 451 |
+
)
|
| 452 |
+
print('====== Below is reasoning content in Thinking Mode ======')
|
| 453 |
+
print(f'reasoning content: {response.choices[0].message.reasoning}')
|
| 454 |
+
print('====== Below is response in Thinking Mode ======')
|
| 455 |
+
print(f'response: {response.choices[0].message.content}')
|
| 456 |
+
|
| 457 |
+
# To use instant mode, pass {"thinking" = {"type":"disabled"}}
|
| 458 |
+
response = client.chat.completions.create(
|
| 459 |
+
model=model_name,
|
| 460 |
+
messages=messages,
|
| 461 |
+
stream=False,
|
| 462 |
+
max_tokens=4096,
|
| 463 |
+
extra_body={'thinking': {'type': 'disabled'}}, # this is for official API
|
| 464 |
+
# extra_body= {'chat_template_kwargs': {"thinking": False}} # this is for vLLM/SGLang
|
| 465 |
+
)
|
| 466 |
+
print('====== Below is response in Instant Mode ======')
|
| 467 |
+
print(f'response: {response.choices[0].message.content}')
|
| 468 |
+
```
|
| 469 |
+
|
| 470 |
+
|
| 471 |
+
### Chat Completion with visual content
|
| 472 |
+
|
| 473 |
+
K2.6 supports Image and Video input.
|
| 474 |
+
|
| 475 |
+
The following example demonstrates how to call K2.6 API with image input:
|
| 476 |
+
|
| 477 |
+
```python
|
| 478 |
+
import openai
|
| 479 |
+
import base64
|
| 480 |
+
import requests
|
| 481 |
+
|
| 482 |
+
def chat_with_image(client: openai.OpenAI, model_name: str):
|
| 483 |
+
url = 'https://huggingface.co/moonshotai/Kimi-K2.6/resolve/main/figures/kimi-logo.png'
|
| 484 |
+
image_base64 = base64.b64encode(requests.get(url).content).decode()
|
| 485 |
+
messages = [
|
| 486 |
+
{
|
| 487 |
+
'role': 'user',
|
| 488 |
+
'content': [
|
| 489 |
+
{'type': 'text', 'text': 'Describe this image in detail.'},
|
| 490 |
+
{
|
| 491 |
+
'type': 'image_url',
|
| 492 |
+
'image_url': {'url': f'data:image/png;base64, {image_base64}'},
|
| 493 |
+
},
|
| 494 |
+
],
|
| 495 |
+
}
|
| 496 |
+
]
|
| 497 |
+
|
| 498 |
+
response = client.chat.completions.create(
|
| 499 |
+
model=model_name, messages=messages, stream=False, max_tokens=8192
|
| 500 |
+
)
|
| 501 |
+
print('====== Below is reasoning content in Thinking Mode ======')
|
| 502 |
+
print(f'reasoning content: {response.choices[0].message.reasoning}')
|
| 503 |
+
print('====== Below is response in Thinking Mode ======')
|
| 504 |
+
print(f'response: {response.choices[0].message.content}')
|
| 505 |
+
|
| 506 |
+
# Also support instant mode if you pass {"thinking" = {"type":"disabled"}}
|
| 507 |
+
response = client.chat.completions.create(
|
| 508 |
+
model=model_name,
|
| 509 |
+
messages=messages,
|
| 510 |
+
stream=False,
|
| 511 |
+
max_tokens=4096,
|
| 512 |
+
extra_body={'thinking': {'type': 'disabled'}}, # this is for official API
|
| 513 |
+
# extra_body= {'chat_template_kwargs': {"thinking": False}} # this is for vLLM/SGLang
|
| 514 |
+
)
|
| 515 |
+
print('====== Below is response in Instant Mode ======')
|
| 516 |
+
print(f'response: {response.choices[0].message.content}')
|
| 517 |
+
|
| 518 |
+
return response.choices[0].message.content
|
| 519 |
+
```
|
| 520 |
+
|
| 521 |
+
The following example demonstrates how to call K2.6 API with video input:
|
| 522 |
+
|
| 523 |
+
```python
|
| 524 |
+
import openai
|
| 525 |
+
import base64
|
| 526 |
+
import requests
|
| 527 |
+
|
| 528 |
+
def chat_with_video(client: openai.OpenAI, model_name:str):
|
| 529 |
+
url = 'https://huggingface.co/moonshotai/Kimi-K2.6/resolve/main/figures/demo_video.mp4'
|
| 530 |
+
video_base64 = base64.b64encode(requests.get(url).content).decode()
|
| 531 |
+
messages = [
|
| 532 |
+
{
|
| 533 |
+
"role": "user",
|
| 534 |
+
"content": [
|
| 535 |
+
{"type": "text","text": "Describe the video in detail."},
|
| 536 |
+
{
|
| 537 |
+
"type": "video_url",
|
| 538 |
+
"video_url": {"url": f"data:video/mp4;base64,{video_base64}"},
|
| 539 |
+
},
|
| 540 |
+
],
|
| 541 |
+
}
|
| 542 |
+
]
|
| 543 |
+
|
| 544 |
+
response = client.chat.completions.create(model=model_name, messages=messages)
|
| 545 |
+
print('====== Below is reasoning content in Thinking Mode ======')
|
| 546 |
+
print(f'reasoning content: {response.choices[0].message.reasoning}')
|
| 547 |
+
print('====== Below is response in Thinking Mode ======')
|
| 548 |
+
print(f'response: {response.choices[0].message.content}')
|
| 549 |
+
|
| 550 |
+
# Also support instant mode if pass {"thinking" = {"type":"disabled"}}
|
| 551 |
+
response = client.chat.completions.create(
|
| 552 |
+
model=model_name,
|
| 553 |
+
messages=messages,
|
| 554 |
+
stream=False,
|
| 555 |
+
max_tokens=4096,
|
| 556 |
+
extra_body={'thinking': {'type': 'disabled'}}, # this is for official API
|
| 557 |
+
# extra_body= {'chat_template_kwargs': {"thinking": False}} # this is for vLLM/SGLang
|
| 558 |
+
)
|
| 559 |
+
print('====== Below is response in Instant Mode ======')
|
| 560 |
+
print(f'response: {response.choices[0].message.content}')
|
| 561 |
+
return response.choices[0].message.content
|
| 562 |
+
```
|
| 563 |
+
|
| 564 |
+
### Preserve Thinking
|
| 565 |
+
Kimi K2.6 supports `preserve_thinking` mode, which retains full reasoning content across multi-turn interactions and enhances performance in coding agent scenarios.
|
| 566 |
+
|
| 567 |
+
This feature is disabled by default. The following example demonstrates how to call K2.6 API in `preserve_thinking` mode:
|
| 568 |
+
|
| 569 |
+
```python
|
| 570 |
+
def chat_with_preserve_thinking(client: openai.OpenAI, model_name: str):
|
| 571 |
+
messages = [
|
| 572 |
+
{
|
| 573 |
+
"role": "user",
|
| 574 |
+
"content": "Tell me three random numbers."
|
| 575 |
+
},
|
| 576 |
+
{
|
| 577 |
+
"role": "assistant",
|
| 578 |
+
"reasoning_content": "I'll start by listing five numbers: 473, 921, 235, 215, 222, and I'll tell you the first three.",
|
| 579 |
+
"content": "473, 921, 235"
|
| 580 |
+
},
|
| 581 |
+
{
|
| 582 |
+
"role": "user",
|
| 583 |
+
"content": "What are the other two numbers you have in mind?"
|
| 584 |
+
}
|
| 585 |
+
]
|
| 586 |
+
|
| 587 |
+
response = client.chat.completions.create(
|
| 588 |
+
model=model_name,
|
| 589 |
+
messages=messages,
|
| 590 |
+
stream=False,
|
| 591 |
+
max_tokens=4096,
|
| 592 |
+
extra_body={'thinking': {'type': 'enabled', 'keep': 'all'}}, # this is for official API
|
| 593 |
+
# extra_body={"chat_template_kwargs": {"thinking":True, "preserve_thinking": True}}, # this is for vLLM/SGLang
|
| 594 |
+
# We recommend enabling preserve_thinking only in think mode.
|
| 595 |
+
)
|
| 596 |
+
# the assistant should mention 215 and 222 that appear in the prior reasoning content
|
| 597 |
+
print(f"response: {response.choices[0].message.reasoning}")
|
| 598 |
+
return response.choices[0].message.content
|
| 599 |
+
|
| 600 |
+
```
|
| 601 |
+
|
| 602 |
+
### Interleaved Thinking and Multi-Step Tool Call
|
| 603 |
+
|
| 604 |
+
K2.6 shares the same design of Interleaved Thinking and Multi-Step Tool Call as K2 Thinking. For usage example, please refer to the [K2 Thinking documentation](https://platform.moonshot.ai/docs/guide/use-kimi-k2-thinking-model#complete-example).
|
| 605 |
+
|
| 606 |
+
### Coding Agent Framework
|
| 607 |
+
|
| 608 |
+
Kimi K2.6 works best with Kimi Code CLI as its agent framework — give it a try at https://www.kimi.com/code.
|
| 609 |
+
|
| 610 |
+
|
| 611 |
+
---
|
| 612 |
+
|
| 613 |
+
## 7. License
|
| 614 |
+
|
| 615 |
+
Both the code repository and the model weights are released under the [Modified MIT License](LICENSE).
|
| 616 |
+
|
| 617 |
+
---
|
| 618 |
+
|
| 619 |
+
## 8. Third Party Notices
|
| 620 |
+
|
| 621 |
+
See [THIRD PARTY NOTICES](THIRD_PARTY_NOTICES.md)
|
| 622 |
+
|
| 623 |
+
---
|
| 624 |
+
|
| 625 |
+
## 9. Contact Us
|
| 626 |
+
|
| 627 |
+
If you have any questions, please reach out at [support@moonshot.ai](mailto:support@moonshot.ai).
|