LittleDesignSolution/Kimi-K2.5
017
1---2tags:3- compressed-tensors4license: other5license_name: modified-mit6library_name: transformers7pipeline_tag: image-text-to-text8paper: arxiv.org/abs/2602.022769---10<div align="center">11 <picture>12 <img src="figures/kimi-logo.png" width="30%" alt="Kimi K2.5">13 </picture>14</div>15<hr>16<div align="center" style="line-height:1">17 <a href="https://www.kimi.com" target="_blank"><img alt="Chat" src="https://img.shields.io/badge/🤖%20Chat-Kimi%20K2.5-ff6b6b?color=1783ff&logoColor=white"/></a>18 <a href="https://github.com/moonshotai/Kimi-K2.5"><img alt="github" src="https://img.shields.io/badge/Github-Kimi%20K2.5-181717?logo=github&color=1783ff&logoColor=white"/></a>19 <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>20</div>21 22<div align="center" style="line-height: 1;">23 <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>24 <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>25 <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>26</div>27<div align="center" style="line-height: 1;">28 <a href="https://huggingface.co/moonshotai/Kimi-K2.5/blob/main/LICENSE"><img alt="License" src="https://img.shields.io/badge/License-Modified_MIT-f5de53?&color=f5de53"/></a>29</div>30<p align="center">31<b>📰 <a href="https://www.kimi.com/blog/kimi-k2-5.html">Tech Blog</a></b> | <b>📄 <a href="https://arxiv.org/abs/2602.02276">Paper</a></b>32</p> 33</p>34 35## 0. Changelog36- 2026.1.29: 37 - The default system prompt might cause confusion to users and unexpected behaviours, so we remove it.38 - The token `<|media_start|>` is incorrect; it has been replaced with `<|media_begin|>` in the chat template.39 40## 1. Model Introduction41 42Kimi K2.5 is an open-source, native multimodal agentic model built through continual pretraining on approximately 15 trillion mixed visual and text tokens atop Kimi-K2-Base. It seamlessly integrates vision and language understanding with advanced agentic capabilities, instant and thinking modes, as well as conversational and agentic paradigms.43 44### Key Features45- **Native Multimodality**: Pre-trained on vision–language tokens, K2.5 excels in visual knowledge, cross-modal reasoning, and agentic tool use grounded in visual inputs.46- **Coding with Vision**: K2.5 generates code from visual specifications (UI designs, video workflows) and autonomously orchestrates tools for visual data processing.47- **Agent Swarm**: K2.5 transitions from single-agent scaling to a self-directed, coordinated swarm-like execution scheme. It decomposes complex tasks into parallel sub-tasks executed by dynamically instantiated, domain-specific agents.48 49## 2. Model Summary50 51<div align="center">52 53 54| | |55|:---:|:---:|56| **Architecture** | Mixture-of-Experts (MoE) |57| **Total Parameters** | 1T |58| **Activated Parameters** | 32B |59| **Number of Layers** (Dense layer included) | 61 |60| **Number of Dense Layers** | 1 |61| **Attention Hidden Dimension** | 7168 |62| **MoE Hidden Dimension** (per Expert) | 2048 |63| **Number of Attention Heads** | 64 |64| **Number of Experts** | 384 |65| **Selected Experts per Token** | 8 |66| **Number of Shared Experts** | 1 |67| **Vocabulary Size** | 160K |68| **Context Length** | 256K |69| **Attention Mechanism** | MLA |70| **Activation Function** | SwiGLU |71| **Vision Encoder** | MoonViT |72| **Parameters of Vision Encoder** | 400M |73</div>74 75## 3. Evaluation Results76 77 78 79<div align="center">80<table>81<thead>82<tr>83<th align="center">Benchmark</th>84<th align="center"><sup>Kimi K2.5<br><sup>(Thinking)</sup></sup></th>85<th align="center"><sup>GPT-5.2 <br><sup>(xhigh)</sup></sup></th>86<th align="center"><sup>Claude 4.5 Opus <br><sup>(Extended Thinking)</sup></sup></th>87<th align="center"><sup>Gemini 3 Pro <br><sup>(High Thinking Level)</sup></sup></th>88<th align="center"><sup>DeepSeek V3.2 <br><sup>(Thinking)</sup></sup></th>89<th align="center"><sup>Qwen3-VL-<br>235B-A22B-<br>Thinking</sup></th>90</tr>91</thead>92<tbody>93<tr>94<td align="center" colspan=8><strong>Reasoning & Knowledge</strong></td>95</tr>96<tr>97<td align="center" style="vertical-align: middle">HLE-Full</td>98<td align="center" style="vertical-align: middle">30.1</td>99<td align="center" style="vertical-align: middle">34.5</td>100<td align="center" style="vertical-align: middle">30.8</td>101<td align="center" style="vertical-align: middle">37.5</td>102<td align="center" style="vertical-align: middle">25.1<sup>†</sup></td>103<td align="center" style="vertical-align: middle">-</td>104</tr>105<tr>106<td align="center" style="vertical-align: middle">HLE-Full<br>(w/ tools)</td>107<td align="center" style="vertical-align: middle">50.2</td>108<td align="center" style="vertical-align: middle">45.5</td>109<td align="center" style="vertical-align: middle">43.2</td>110<td align="center" style="vertical-align: middle">45.8</td>111<td align="center" style="vertical-align: middle">40.8<sup>†</sup></td>112<td align="center" style="vertical-align: middle">-</td>113</tr>114<tr>115<td align="center" style="vertical-align: middle">AIME 2025</td>116<td align="center" style="vertical-align: middle">96.1</td>117<td align="center" style="vertical-align: middle">100</td>118<td align="center" style="vertical-align: middle">92.8</td>119<td align="center" style="vertical-align: middle">95.0</td>120<td align="center" style="vertical-align: middle">93.1</td>121<td align="center" style="vertical-align: middle">-</td>122</tr>123<tr>124<td align="center" style="vertical-align: middle">HMMT 2025 (Feb)</td>125<td align="center" style="vertical-align: middle">95.4</td>126<td align="center" style="vertical-align: middle">99.4</td>127<td align="center" style="vertical-align: middle">92.9*</td>128<td align="center" style="vertical-align: middle">97.3*</td>129<td align="center" style="vertical-align: middle">92.5</td>130<td align="center" style="vertical-align: middle">-</td>131</tr>132<tr>133<td align="center" style="vertical-align: middle">IMO-AnswerBench</td>134<td align="center" style="vertical-align: middle">81.8</td>135<td align="center" style="vertical-align: middle">86.3</td>136<td align="center" style="vertical-align: middle">78.5*</td>137<td align="center" style="vertical-align: middle">83.1*</td>138<td align="center" style="vertical-align: middle">78.3</td>139<td align="center" style="vertical-align: middle">-</td>140</tr>141<tr>142<td align="center" style="vertical-align: middle">GPQA-Diamond</td>143<td align="center" style="vertical-align: middle">87.6</td>144<td align="center" style="vertical-align: middle">92.4</td>145<td align="center" style="vertical-align: middle">87.0</td>146<td align="center" style="vertical-align: middle">91.9</td>147<td align="center" style="vertical-align: middle">82.4</td>148<td align="center" style="vertical-align: middle">-</td>149</tr>150<tr>151<td align="center" style="vertical-align: middle">MMLU-Pro</td>152<td align="center" style="vertical-align: middle">87.1</td>153<td align="center" style="vertical-align: middle">86.7*</td>154<td align="center" style="vertical-align: middle">89.3*</td>155<td align="center" style="vertical-align: middle">90.1</td>156<td align="center" style="vertical-align: middle">85.0</td>157<td align="center" style="vertical-align: middle">-</td>158</tr>159<tr>160<td align="center" colspan=8><strong>Image & Video</strong></td>161</tr>162<tr>163<td align="center" style="vertical-align: middle">MMMU-Pro</td>164<td align="center" style="vertical-align: middle">78.5</td>165<td align="center" style="vertical-align: middle">79.5*</td>166<td align="center" style="vertical-align: middle">74.0</td>167<td align="center" style="vertical-align: middle">81.0</td>168<td align="center" style="vertical-align: middle">-</td>169<td align="center" style="vertical-align: middle">69.3</td>170</tr>171<tr>172<td align="center" style="vertical-align: middle">CharXiv (RQ)</td>173<td align="center" style="vertical-align: middle">77.5</td>174<td align="center" style="vertical-align: middle">82.1</td>175<td align="center" style="vertical-align: middle">67.2*</td>176<td align="center" style="vertical-align: middle">81.4</td>177<td align="center" style="vertical-align: middle">-</td>178<td align="center" style="vertical-align: middle">66.1</td>179</tr>180<tr>181<td align="center" style="vertical-align: middle">MathVision</td>182<td align="center" style="vertical-align: middle">84.2</td>183<td align="center" style="vertical-align: middle">83.0</td>184<td align="center" style="vertical-align: middle">77.1*</td>185<td align="center" style="vertical-align: middle">86.1*</td>186<td align="center" style="vertical-align: middle">-</td>187<td align="center" style="vertical-align: middle">74.6</td>188</tr>189<tr>190<td align="center" style="vertical-align: middle">MathVista (mini)</td>191<td align="center" style="vertical-align: middle">90.1</td>192<td align="center" style="vertical-align: middle">82.8*</td>193<td align="center" style="vertical-align: middle">80.2*</td>194<td align="center" style="vertical-align: middle">89.8*</td>195<td align="center" style="vertical-align: middle">-</td>196<td align="center" style="vertical-align: middle">85.8</td>197</tr>198<tr>199<td align="center" style="vertical-align: middle">ZeroBench</td>200<td align="center" style="vertical-align: middle">9</td>201<td align="center" style="vertical-align: middle">9*</td>202<td align="center" style="vertical-align: middle">3*</td>203<td align="center" style="vertical-align: middle">8*</td>204<td align="center" style="vertical-align: middle">-</td>205<td align="center" style="vertical-align: middle">4*</td>206</tr>207<tr>208<td align="center" style="vertical-align: middle">ZeroBench<br>(w/ tools)</td>209<td align="center" style="vertical-align: middle">11</td>210<td align="center" style="vertical-align: middle">7*</td>211<td align="center" style="vertical-align: middle">9*</td>212<td align="center" style="vertical-align: middle">12*</td>213<td align="center" style="vertical-align: middle">-</td>214<td align="center" style="vertical-align: middle">3*</td>215</tr>216<tr>217<td align="center" style="vertical-align: middle">OCRBench</td>218<td align="center" style="vertical-align: middle">92.3</td>219<td align="center" style="vertical-align: middle">80.7*</td>220<td align="center" style="vertical-align: middle">86.5*</td>221<td align="center" style="vertical-align: middle">90.3*</td>222<td align="center" style="vertical-align: middle">-</td>223<td align="center" style="vertical-align: middle">87.5</td>224</tr>225<tr>226<td align="center" style="vertical-align: middle">OmniDocBench 1.5</td>227<td align="center" style="vertical-align: middle">88.8</td>228<td align="center" style="vertical-align: middle">85.7</td>229<td align="center" style="vertical-align: middle">87.7*</td>230<td align="center" style="vertical-align: middle">88.5</td>231<td align="center" style="vertical-align: middle">-</td>232<td align="center" style="vertical-align: middle">82.0*</td>233</tr>234<tr>235<td align="center" style="vertical-align: middle">InfoVQA (val)</td>236<td align="center" style="vertical-align: middle">92.6</td>237<td align="center" style="vertical-align: middle">84*</td>238<td align="center" style="vertical-align: middle">76.9*</td>239<td align="center" style="vertical-align: middle">57.2*</td>240<td align="center" style="vertical-align: middle">-</td>241<td align="center" style="vertical-align: middle">89.5</td>242</tr>243<tr>244<td align="center" style="vertical-align: middle">SimpleVQA</td>245<td align="center" style="vertical-align: middle">71.2</td>246<td align="center" style="vertical-align: middle">55.8*</td>247<td align="center" style="vertical-align: middle">69.7*</td>248<td align="center" style="vertical-align: middle">69.7*</td>249<td align="center" style="vertical-align: middle">-</td>250<td align="center" style="vertical-align: middle">56.8*</td>251</tr>252<tr>253<td align="center" style="vertical-align: middle"><a href="https://github.com/MoonshotAI/WorldVQA">WorldVQA</a></td>254<td align="center" style="vertical-align: middle">46.3</td>255<td align="center" style="vertical-align: middle">28.0</td>256<td align="center" style="vertical-align: middle">36.8</td>257<td align="center" style="vertical-align: middle">47.4</td>258<td align="center" style="vertical-align: middle">-</td>259<td align="center" style="vertical-align: middle">23.5</td>260</tr>261<tr>262<td align="center" style="vertical-align: middle">VideoMMMU</td>263<td align="center" style="vertical-align: middle">86.6</td>264<td align="center" style="vertical-align: middle">85.9</td>265<td align="center" style="vertical-align: middle">84.4*</td>266<td align="center" style="vertical-align: middle">87.6</td>267<td align="center" style="vertical-align: middle">-</td>268<td align="center" style="vertical-align: middle">80.0</td>269</tr>270<tr>271<td align="center" style="vertical-align: middle">MMVU</td>272<td align="center" style="vertical-align: middle">80.4</td>273<td align="center" style="vertical-align: middle">80.8*</td>274<td align="center" style="vertical-align: middle">77.3</td>275<td align="center" style="vertical-align: middle">77.5</td>276<td align="center" style="vertical-align: middle">-</td>277<td align="center" style="vertical-align: middle">71.1</td>278</tr>279<tr>280<td align="center" style="vertical-align: middle">MotionBench</td>281<td align="center" style="vertical-align: middle">70.4</td>282<td align="center" style="vertical-align: middle">64.8</td>283<td align="center" style="vertical-align: middle">60.3</td>284<td align="center" style="vertical-align: middle">70.3</td>285<td align="center" style="vertical-align: middle">-</td>286<td align="center" style="vertical-align: middle">-</td>287</tr>288<tr>289<td align="center" style="vertical-align: middle">VideoMME</td>290<td align="center" style="vertical-align: middle">87.4</td>291<td align="center" style="vertical-align: middle">86.0*</td>292<td align="center" style="vertical-align: middle">-</td>293<td align="center" style="vertical-align: middle">88.4*</td>294<td align="center" style="vertical-align: middle">-</td>295<td align="center" style="vertical-align: middle">79.0</td>296</tr>297<tr>298<td align="center" style="vertical-align: middle">LongVideoBench</td>299<td align="center" style="vertical-align: middle">79.8</td>300<td align="center" style="vertical-align: middle">76.5*</td>301<td align="center" style="vertical-align: middle">67.2*</td>302<td align="center" style="vertical-align: middle">77.7*</td>303<td align="center" style="vertical-align: middle">-</td>304<td align="center" style="vertical-align: middle">65.6*</td>305</tr>306<tr>307<td align="center" style="vertical-align: middle">LVBench</td>308<td align="center" style="vertical-align: middle">75.9</td>309<td align="center" style="vertical-align: middle">-</td>310<td align="center" style="vertical-align: middle">-</td>311<td align="center" style="vertical-align: middle">73.5*</td>312<td align="center" style="vertical-align: middle">-</td>313<td align="center" style="vertical-align: middle">63.6</td>314</tr>315<tr>316<td align="center" colspan=8><strong>Coding</strong></td>317</tr>318<tr>319<td align="center" style="vertical-align: middle">SWE-Bench Verified</td>320<td align="center" style="vertical-align: middle">76.8</td>321<td align="center" style="vertical-align: middle">80.0</td>322<td align="center" style="vertical-align: middle">80.9</td>323<td align="center" style="vertical-align: middle">76.2</td>324<td align="center" style="vertical-align: middle">73.1</td>325<td align="center" style="vertical-align: middle">-</td>326</tr>327<tr>328<td align="center" style="vertical-align: middle">SWE-Bench Pro</td>329<td align="center" style="vertical-align: middle">50.7</td>330<td align="center" style="vertical-align: middle">55.6</td>331<td align="center" style="vertical-align: middle">55.4*</td>332<td align="center" style="vertical-align: middle">-</td>333<td align="center" style="vertical-align: middle">-</td>334<td align="center" style="vertical-align: middle">-</td>335</tr>336<tr>337<td align="center" style="vertical-align: middle">SWE-Bench Multilingual</td>338<td align="center" style="vertical-align: middle">73.0</td>339<td align="center" style="vertical-align: middle">72.0</td>340<td align="center" style="vertical-align: middle">77.5</td>341<td align="center" style="vertical-align: middle">65.0</td>342<td align="center" style="vertical-align: middle">70.2</td>343<td align="center" style="vertical-align: middle">-</td>344</tr>345<tr>346<td align="center" style="vertical-align: middle">Terminal Bench 2.0</td>347<td align="center" style="vertical-align: middle">50.8</td>348<td align="center" style="vertical-align: middle">54.0</td>349<td align="center" style="vertical-align: middle">59.3</td>350<td align="center" style="vertical-align: middle">54.2</td>351<td align="center" style="vertical-align: middle">46.4</td>352<td align="center" style="vertical-align: middle">-</td>353</tr>354<tr>355<td align="center" style="vertical-align: middle">PaperBench</td>356<td align="center" style="vertical-align: middle">63.5</td>357<td align="center" style="vertical-align: middle">63.7*</td>358<td align="center" style="vertical-align: middle">72.9*</td>359<td align="center" style="vertical-align: middle">-</td>360<td align="center" style="vertical-align: middle">47.1</td>361<td align="center" style="vertical-align: middle">-</td>362</tr>363<tr>364<td align="center" style="vertical-align: middle">CyberGym</td>365<td align="center" style="vertical-align: middle">41.3</td>366<td align="center" style="vertical-align: middle">-</td>367<td align="center" style="vertical-align: middle">50.6</td>368<td align="center" style="vertical-align: middle">39.9*</td>369<td align="center" style="vertical-align: middle">17.3*</td>370<td align="center" style="vertical-align: middle">-</td>371</tr>372<tr>373<td align="center" style="vertical-align: middle">SciCode</td>374<td align="center" style="vertical-align: middle">48.7</td>375<td align="center" style="vertical-align: middle">52.1</td>376<td align="center" style="vertical-align: middle">49.5</td>377<td align="center" style="vertical-align: middle">56.1</td>378<td align="center" style="vertical-align: middle">38.9</td>379<td align="center" style="vertical-align: middle">-</td>380</tr>381<tr>382<td align="center" style="vertical-align: middle">OJBench (cpp)</td>383<td align="center" style="vertical-align: middle">57.4</td>384<td align="center" style="vertical-align: middle">-</td>385<td align="center" style="vertical-align: middle">54.6*</td>386<td align="center" style="vertical-align: middle">68.5*</td>387<td align="center" style="vertical-align: middle">54.7*</td>388<td align="center" style="vertical-align: middle">-</td>389</tr>390<tr>391<td align="center" style="vertical-align: middle">LiveCodeBench (v6)</td>392<td align="center" style="vertical-align: middle">85.0</td>393<td align="center" style="vertical-align: middle">-</td>394<td align="center" style="vertical-align: middle">82.2*</td>395<td align="center" style="vertical-align: middle">87.4*</td>396<td align="center" style="vertical-align: middle">83.3</td>397<td align="center" style="vertical-align: middle">-</td>398</tr>399<tr>400<td align="center" colspan=8><strong>Long Context</strong></td>401</tr>402<tr>403<td align="center" style="vertical-align: middle">Longbench v2</td>404<td align="center" style="vertical-align: middle">61.0</td>405<td align="center" style="vertical-align: middle">54.5*</td>406<td align="center" style="vertical-align: middle">64.4*</td>407<td align="center" style="vertical-align: middle">68.2*</td>408<td align="center" style="vertical-align: middle">59.8*</td>409<td align="center" style="vertical-align: middle">-</td>410</tr>411<tr>412<td align="center" style="vertical-align: middle">AA-LCR</td>413<td align="center" style="vertical-align: middle">70.0</td>414<td align="center" style="vertical-align: middle">72.3*</td>415<td align="center" style="vertical-align: middle">71.3*</td>416<td align="center" style="vertical-align: middle">65.3*</td>417<td align="center" style="vertical-align: middle">64.3*</td>418<td align="center" style="vertical-align: middle">-</td>419<tr>420<td align="center" colspan=8><strong>Agentic Search</strong></td>421</tr>422<tr>423<td align="center" style="vertical-align: middle">BrowseComp</td>424<td align="center" style="vertical-align: middle">60.6</td>425<td align="center" style="vertical-align: middle" rowspan="2">65.8</td>426<td align="center" style="vertical-align: middle">37.0</td>427<td align="center" style="vertical-align: middle">37.8</td>428<td align="center" style="vertical-align: middle">51.4</td>429<td align="center" style="vertical-align: middle">-</td>430</tr>431<tr>432<td align="center" style="vertical-align: middle">BrowseComp<br>(w/ctx manage)</td>433<td align="center" style="vertical-align: middle">74.9</td>434<td align="center" style="vertical-align: middle">57.8</td>435<td align="center" style="vertical-align: middle">59.2</td>436<td align="center" style="vertical-align: middle">67.6</td>437<td align="center" style="vertical-align: middle">-</td>438</tr>439<tr>440<td align="center" style="vertical-align: middle">BrowseComp<br>(Agent Swarm)</td>441<td align="center" style="vertical-align: middle">78.4</td>442<td align="center" style="vertical-align: middle">-</td>443<td align="center" style="vertical-align: middle">-</td>444<td align="center" style="vertical-align: middle">-</td>445<td align="center" style="vertical-align: middle">-</td>446<td align="center" style="vertical-align: middle">-</td>447</tr>448<tr>449<td align="center" style="vertical-align: middle">WideSearch<br> (item-f1)</td>450<td align="center" style="vertical-align: middle">72.7</td>451<td align="center" style="vertical-align: middle">-</td>452<td align="center" style="vertical-align: middle">76.2*</td>453<td align="center" style="vertical-align: middle">57.0</td>454<td align="center" style="vertical-align: middle">32.5*</td>455<td align="center" style="vertical-align: middle">-</td>456</tr>457<tr>458<td align="center" style="vertical-align: middle">WideSearch<br> (item-f1 Agent Swarm)</td>459<td align="center" style="vertical-align: middle">79.0</td>460<td align="center" style="vertical-align: middle">-</td>461<td align="center" style="vertical-align: middle">-</td>462<td align="center" style="vertical-align: middle">-</td>463<td align="center" style="vertical-align: middle">-</td>464<td align="center" style="vertical-align: middle">-</td>465</tr>466<tr>467<td align="center" style="vertical-align: middle">DeepSearchQA</td>468<td align="center" style="vertical-align: middle">77.1</td>469<td align="center" style="vertical-align: middle">71.3*</td>470<td align="center" style="vertical-align: middle">76.1*</td>471<td align="center" style="vertical-align: middle">63.2*</td>472<td align="center" style="vertical-align: middle">60.9*</td>473<td align="center" style="vertical-align: middle">-</td>474</tr>475<tr>476<td align="center" style="vertical-align: middle">FinSearchCompT2&T3</td>477<td align="center" style="vertical-align: middle">67.8</td>478<td align="center" style="vertical-align: middle">-</td>479<td align="center" style="vertical-align: middle">66.2*</td>480<td align="center" style="vertical-align: middle">49.9</td>481<td align="center" style="vertical-align: middle">59.1*</td>482<td align="center" style="vertical-align: middle">-</td>483</tr>484<tr>485<td align="center" style="vertical-align: middle">Seal-0</td>486<td align="center" style="vertical-align: middle">57.4</td>487<td align="center" style="vertical-align: middle">45.0</td>488<td align="center" style="vertical-align: middle">47.7*</td>489<td align="center" style="vertical-align: middle">45.5*</td>490<td align="center" style="vertical-align: middle">49.5*</td>491<td align="center" style="vertical-align: middle">-</td>492</tr>493</tbody>494</table>495</div>496 497<details>498<summary><b>Footnotes</b></summary>499 5001. General Testing Details501 - We report results for Kimi K2.5 and DeepSeek-V3.2 with thinking mode enabled, Claude Opus 4.5 with extended thinking mode, GPT-5.2 with xhigh reasoning effort, and Gemini 3 Pro with a high thinking level. For vision benchmarks, we additionally report results for Qwen3-VL-235B-A22B-Thinking.502 - Unless otherwise specified, all Kimi K2.5 experiments were conducted with temperature = 1.0, top-p = 0.95, and a context length of 256k tokens.503 - Benchmarks without publicly available scores were re-evaluated under the same conditions used for Kimi K2.5 and are marked with an asterisk (*).504 - We could not evaluate GPT-5.2 xhigh on all benchmarks due to service stability issues. For benchmarks that were not tested, we mark them as "-".5052. Text and Reasoning506 - HLE, AIME 2025, HMMT 2025 (Feb), and GPQA-Diamond were evaluated with a maximum completion budget of 96k tokens.507 - Results for AIME and HMMT are averaged over 32 runs (avg@32); GPQA-Diamond over 8 runs (avg@8).508 - For HLE, we report scores on the full set (text & image). Kimi K2.5 scores 31.5 (text) and 21.3 (image) without tools, and 51.8 (text) and 39.8 (image) with tools. The DeepSeek-V3.2 score corresponds to its text-only subset (marked with †) . Hugging Face access was blocked to prevent potential data leakage. HLE with tools uses simple context management: once the context exceeds a threshold, only the latest round of tool messages is retained.5093. Tool-Augmented / Agentic Search510 - Kimi K2.5 was equipped with search, code-interpreter, and web-browsing tools for HLE with tools and all agentic search benchmarks.511 - Except for BrowseComp (where K2.5 and DeepSeek-V3.2 used the discard-all strategy), no context management was applied, and tasks exceeding the supported context length were directly counted as failed.512 - The test system prompts emphasize deep and proactive tool use, instructing models to reason carefully, leverage tools, and verify uncertain information. Full prompts will be provided in the technical report.513 - Results for Seal-0 and WideSearch are averaged over four runs (avg@4).5144. Vision Benchmarks515 - Max-tokens = 64k, averaged over three runs (avg@3).516 - ZeroBench (w/ tools) uses max-tokens-per-step = 24k and max-steps = 30 for multi-step reasoning.517 - MMMU-Pro follows the official protocol, preserving input order and prepending images.518 - GPT-5.2-xhigh had ~10% failure rate (no output despite 3 retries), treated as incorrect; reported scores likely underestimate true performance.519 - WorldVQA, a benchmark designed to evaluate atomic vision-centric world knowledge. Access WorldVQA at https://github.com/MoonshotAI/WorldVQA.520 - OmniDocBench Score is computed as (1 − normalized Levenshtein distance) × 100, where a higher score denotes superior accuracy.5215. Coding Tasks522 - Terminal-Bench 2.0 scores were obtained with the default agent framework (Terminus-2) and the provided JSON parser. In our implementation, we evaluated Terminal-Bench 2.0 under non-thinking mode. This choice was made because our current context management strategy for the thinking mode is incompatible with Terminus-2.523 - For the SWE-Bench series of evaluations (including verified, multilingual, and pro), we used an internally developed evaluation framework. This framework includes a minimal set of tools—bash tool, createfile tool, insert tool, view tool, strreplace tool, and submit tool—along with tailored system prompts designed for the tasks. The highest scores were achieved under non-thinking mode.524 - The score of Claude Opus 4.5 on CyberGym is reported under the non-thinking setting.525 - All reported scores of coding tasks are averaged over 5 independent runs.5266. Long-Context Benchmarks527 - AA-LCR: scores averaged over three runs (avg@3).528 - LongBench-V2: identical prompts and input contexts standardized to ~128k tokens.5297. Agent Swarm530 - BrowseComp (Swarm Mode): main agent max 15 steps; sub-agents max 100 steps.531 - WideSearch (Swarm Mode): main and sub-agents max 100 steps.532 533</details>534 535## 4. Native INT4 Quantization536Kimi-K2.5 adopts the same native int4 quantization method as [Kimi-K2-Thinking](https://huggingface.co/moonshotai/Kimi-K2-Thinking#4-native-int4-quantization).537 538## 5. Deployment539> [!Note]540> You can access Kimi-K2.5'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).541Currently, Kimi-K2.5 is recommended to run on the following inference engines:542* vLLM543* SGLang544* KTransformers545 546The minimum version requirement for `transformers` is `4.57.1`.547 548Deployment examples can be found in the [Model Deployment Guide](docs/deploy_guidance.md).549 550 551---552## 6. Model Usage553 554The usage demos below demonstrate how to call our official API. 555 556For third-party APIs deployed with vLLM or SGLang, please note that:557> [!Note]558> - Chat with video content is an experimental feature and is only supported in our official API for now.559> 560> - The recommended `temperature` will be `1.0` for Thinking mode and `0.6` for Instant mode.561> 562> - The recommended `top_p` is `0.95`.563> 564> - To use instant mode, you need to pass `{'chat_template_kwargs': {"thinking": False}}` in `extra_body`.565 566### Chat Completion567 568This is a simple chat completion script which shows how to call K2.5 API in Thinking and Instant modes.569 570```python571import openai572import base64573import requests574def simple_chat(client: openai.OpenAI, model_name: str):575 messages = [576 {'role': 'system', 'content': 'You are Kimi, an AI assistant created by Moonshot AI.'},577 {578 'role': 'user',579 'content': [580 {'type': 'text', 'text': 'which one is bigger, 9.11 or 9.9? think carefully.'}581 ],582 },583 ]584 response = client.chat.completions.create(585 model=model_name, messages=messages, stream=False, max_tokens=4096586 )587 print('====== Below is reasoning_content in Thinking Mode ======')588 print(f'reasoning content: {response.choices[0].message.reasoning_content}')589 print('====== Below is response in Thinking Mode ======')590 print(f'response: {response.choices[0].message.content}')591 592 # To use instant mode, pass {"thinking" = {"type":"disabled"}}593 response = client.chat.completions.create(594 model=model_name,595 messages=messages,596 stream=False,597 max_tokens=4096,598 extra_body={'thinking': {'type': 'disabled'}}, # this is for official API599 # extra_body= {'chat_template_kwargs': {"thinking": False}} # this is for vLLM/SGLang600 )601 print('====== Below is response in Instant Mode ======')602 print(f'response: {response.choices[0].message.content}')603```604 605 606### Chat Completion with visual content607 608K2.5 supports Image and Video input. 609 610The following example demonstrates how to call K2.5 API with image input:611 612```python613import openai614import base64615import requests616 617def chat_with_image(client: openai.OpenAI, model_name: str):618 url = 'https://huggingface.co/moonshotai/Kimi-K2.5/resolve/main/figures/kimi-logo.png'619 image_base64 = base64.b64encode(requests.get(url).content).decode()620 messages = [621 {622 'role': 'user',623 'content': [624 {625 'type': 'image_url',626 'image_url': {'url': f'data:image/png;base64, {image_base64}'},627 },628 {'type': 'text', 'text': 'Describe this image in detail.'},629 ],630 }631 ]632 633 response = client.chat.completions.create(634 model=model_name, messages=messages, stream=False, max_tokens=8192635 )636 print('====== Below is reasoning_content in Thinking Mode ======')637 print(f'reasoning content: {response.choices[0].message.reasoning_content}')638 print('====== Below is response in Thinking Mode ======')639 print(f'response: {response.choices[0].message.content}')640 641 # Also support instant mode if you pass {"thinking" = {"type":"disabled"}}642 response = client.chat.completions.create(643 model=model_name,644 messages=messages,645 stream=False,646 max_tokens=4096,647 extra_body={'thinking': {'type': 'disabled'}}, # this is for official API648 # extra_body= {'chat_template_kwargs': {"thinking": False}} # this is for vLLM/SGLang649 )650 print('====== Below is response in Instant Mode ======')651 print(f'response: {response.choices[0].message.content}')652 653 return response.choices[0].message.content654```655 656The following example demonstrates how to call K2.5 API with video input:657 658```python659import openai660import base64661import requests662 663def chat_with_video(client: openai.OpenAI, model_name:str):664 url = 'https://huggingface.co/moonshotai/Kimi-K2.5/resolve/main/figures/demo_video.mp4'665 video_base64 = base64.b64encode(requests.get(url).content).decode()666 messages = [667 {668 "role": "user",669 "content": [670 {671 "type": "video_url",672 "video_url": {"url": f"data:video/mp4;base64,{video_base64}"},673 },674 {"type": "text","text": "Describe the video in detail."},675 ],676 }677 ]678 679 response = client.chat.completions.create(model=model_name, messages=messages)680 print('====== Below is reasoning_content in Thinking Mode ======')681 print(f'reasoning content: {response.choices[0].message.reasoning_content}')682 print('====== Below is response in Thinking Mode ======')683 print(f'response: {response.choices[0].message.content}')684 685 # Also support instant mode if pass {"thinking" = {"type":"disabled"}}686 response = client.chat.completions.create(687 model=model_name,688 messages=messages,689 stream=False,690 max_tokens=4096,691 extra_body={'thinking': {'type': 'disabled'}}, # this is for official API692 # extra_body= {'chat_template_kwargs': {"thinking": False}} # this is for vLLM/SGLang693 )694 print('====== Below is response in Instant Mode ======')695 print(f'response: {response.choices[0].message.content}')696 return response.choices[0].message.content697```698 699### Interleaved Thinking and Multi-Step Tool Call700 701K2.5 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).702 703 704### Coding Agent Framework705 706Kimi K2.5 works best with Kimi Code CLI as its agent framework — give it a try at https://www.kimi.com/code.707 708 709---710 711## 7. License712 713Both the code repository and the model weights are released under the [Modified MIT License](LICENSE).714 715---716 717## 8. Third Party Notices718 719See [THIRD PARTY NOTICES](THIRD_PARTY_NOTICES.md)720 721---722 723## 9. Contact Us724 725If you have any questions, please reach out at [support@moonshot.cn](mailto:support@moonshot.cn).726 727## 10. Reference728 729If you find K2.5 useful for your research, please kindly cite K2.5 technical report as follows:730 731```bibtex732@misc{kimiteam2026kimik25visualagentic,733 title={Kimi K2.5: Visual Agentic Intelligence}, 734 author={Kimi Team and Tongtong Bai and Yifan Bai and Yiping Bao and S. H. Cai and Yuan Cao and Y. Charles and H. S. Che and Cheng Chen and Guanduo Chen and Huarong Chen and Jia Chen and Jiahao Chen and Jianlong Chen and Jun Chen and Kefan Chen and Liang Chen and Ruijue Chen and Xinhao Chen and Yanru Chen and Yanxu Chen and Yicun Chen and Yimin Chen and Yingjiang Chen and Yuankun Chen and Yujie Chen and Yutian Chen and Zhirong Chen and Ziwei Chen and Dazhi Cheng and Minghan Chu and Jialei Cui and Jiaqi Deng and Muxi Diao and Hao Ding and Mengfan Dong and Mengnan Dong and Yuxin Dong and Yuhao Dong and Angang Du and Chenzhuang Du and Dikang Du and Lingxiao Du and Yulun Du and Yu Fan and Shengjun Fang and Qiulin Feng and Yichen Feng and Garimugai Fu and Kelin Fu and Hongcheng Gao and Tong Gao and Yuyao Ge and Shangyi Geng and Chengyang Gong and Xiaochen Gong and Zhuoma Gongque and Qizheng Gu and Xinran Gu and Yicheng Gu and Longyu Guan and Yuanying Guo and Xiaoru Hao and Weiran He and Wenyang He and Yunjia He and Chao Hong and Hao Hu and Jiaxi Hu and Yangyang Hu and Zhenxing Hu and Ke Huang and Ruiyuan Huang and Weixiao Huang and Zhiqi Huang and Tao Jiang and Zhejun Jiang and Xinyi Jin and Yu Jing and Guokun Lai and Aidi Li and C. Li and Cheng Li and Fang Li and Guanghe Li and Guanyu Li and Haitao Li and Haoyang Li and Jia Li and Jingwei Li and Junxiong Li and Lincan Li and Mo Li and Weihong Li and Wentao Li and Xinhang Li and Xinhao Li and Yang Li and Yanhao Li and Yiwei Li and Yuxiao Li and Zhaowei Li and Zheming Li and Weilong Liao and Jiawei Lin and Xiaohan Lin and Zhishan Lin and Zichao Lin and Cheng Liu and Chenyu Liu and Hongzhang Liu and Liang Liu and Shaowei Liu and Shudong Liu and Shuran Liu and Tianwei Liu and Tianyu Liu and Weizhou Liu and Xiangyan Liu and Yangyang Liu and Yanming Liu and Yibo Liu and Yuanxin Liu and Yue Liu and Zhengying Liu and Zhongnuo Liu and Enzhe Lu and Haoyu Lu and Zhiyuan Lu and Junyu Luo and Tongxu Luo and Yashuo Luo and Long Ma and Yingwei Ma and Shaoguang Mao and Yuan Mei and Xin Men and Fanqing Meng and Zhiyong Meng and Yibo Miao and Minqing Ni and Kun Ouyang and Siyuan Pan and Bo Pang and Yuchao Qian and Ruoyu Qin and Zeyu Qin and Jiezhong Qiu and Bowen Qu and Zeyu Shang and Youbo Shao and Tianxiao Shen and Zhennan Shen and Juanfeng Shi and Lidong Shi and Shengyuan Shi and Feifan Song and Pengwei Song and Tianhui Song and Xiaoxi Song and Hongjin Su and Jianlin Su and Zhaochen Su and Lin Sui and Jinsong Sun and Junyao Sun and Tongyu Sun and Flood Sung and Yunpeng Tai and Chuning Tang and Heyi Tang and Xiaojuan Tang and Zhengyang Tang and Jiawen Tao and Shiyuan Teng and Chaoran Tian and Pengfei Tian and Ao Wang and Bowen Wang and Chensi Wang and Chuang Wang and Congcong Wang and Dingkun Wang and Dinglu Wang and Dongliang Wang and Feng Wang and Hailong Wang and Haiming Wang and Hengzhi Wang and Huaqing Wang and Hui Wang and Jiahao Wang and Jinhong Wang and Jiuzheng Wang and Kaixin Wang and Linian Wang and Qibin Wang and Shengjie Wang and Shuyi Wang and Si Wang and Wei Wang and Xiaochen Wang and Xinyuan Wang and Yao Wang and Yejie Wang and Yipu Wang and Yiqin Wang and Yucheng Wang and Yuzhi Wang and Zhaoji Wang and Zhaowei Wang and Zhengtao Wang and Zhexu Wang and Zihan Wang and Zizhe Wang and Chu Wei and Ming Wei and Chuan Wen and Zichen Wen and Chengjie Wu and Haoning Wu and Junyan Wu and Rucong Wu and Wenhao Wu and Yuefeng Wu and Yuhao Wu and Yuxin Wu and Zijian Wu and Chenjun Xiao and Jin Xie and Xiaotong Xie and Yuchong Xie and Yifei Xin and Bowei Xing and Boyu Xu and Jianfan Xu and Jing Xu and Jinjing Xu and L. H. Xu and Lin Xu and Suting Xu and Weixin Xu and Xinbo Xu and Xinran Xu and Yangchuan Xu and Yichang Xu and Yuemeng Xu and Zelai Xu and Ziyao Xu and Junjie Yan and Yuzi Yan and Guangyao Yang and Hao Yang and Junwei Yang and Kai Yang and Ningyuan Yang and Ruihan Yang and Xiaofei Yang and Xinlong Yang and Ying Yang and Yi Yang and Yi Yang and Zhen Yang and Zhilin Yang and Zonghan Yang and Haotian Yao and Dan Ye and Wenjie Ye and Zhuorui Ye and Bohong Yin and Chengzhen Yu and Longhui Yu and Tao Yu and Tianxiang Yu and Enming Yuan and Mengjie Yuan and Xiaokun Yuan and Yang Yue and Weihao Zeng and Dunyuan Zha and Haobing Zhan and Dehao Zhang and Hao Zhang and Jin Zhang and Puqi Zhang and Qiao Zhang and Rui Zhang and Xiaobin Zhang and Y. Zhang and Yadong Zhang and Yangkun Zhang and Yichi Zhang and Yizhi Zhang and Yongting Zhang and Yu Zhang and Yushun Zhang and Yutao Zhang and Yutong Zhang and Zheng Zhang and Chenguang Zhao and Feifan Zhao and Jinxiang Zhao and Shuai Zhao and Xiangyu Zhao and Yikai Zhao and Zijia Zhao and Huabin Zheng and Ruihan Zheng and Shaojie Zheng and Tengyang Zheng and Junfeng Zhong and Longguang Zhong and Weiming Zhong and M. Zhou and Runjie Zhou and Xinyu Zhou and Zaida Zhou and Jinguo Zhu and Liya Zhu and Xinhao Zhu and Yuxuan Zhu and Zhen Zhu and Jingze Zhuang and Weiyu Zhuang and Ying Zou and Xinxing Zu},735 year={2026},736 eprint={2602.02276},737 archivePrefix={arXiv},738 primaryClass={cs.CL},739 url={https://arxiv.org/abs/2602.02276}, 740}741```742