unsloth/Kimi-K2.6-GGUF · Hugging Face

Source: https://huggingface.co/unsloth/Kimi-K2.6-GGUF

https://huggingface.co/unsloth/Kimi-K2.6-GGUF#read-our-how-to-run-kimi-k26-guideRead our How toRun Kimi K2.6 Guide!

SeeUnsloth Dynamic 2.0 GGUFsfor our quantization benchmarks.

• To run Kimi K2.6 in full precision, run Q8 (UD-Q8_K_XL), which is only 10GB bigger than Q4 (UD-Q4_K_XL).
• Smaller GGUFs are still uploading.

UD\-Q8\_K\_XLis lossless from Kimi-K2.6. Kimi uses int4 for MoE weights & BF16 rest. We follow that forQ8\_K\_XL UD\-Q4\_K\_XLis the same asUD\-Q8\_K\_XLexcept all other tensors are Q8_0. BF16is provided as an artifact -UD\-Q8\_K\_XLis fully un-quantized, so use that.

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.

https://huggingface.co/unsloth/Kimi-K2.6-GGUF#key-featuresKey Features

• 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.
• 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.
• 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.
• 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.

https://huggingface.co/unsloth/Kimi-K2.6-GGUF#2-model-summary2. Model Summary

ArchitectureMixture-of-Experts (MoE)Total Parameters1TActivated Parameters32BNumber of Layers(Dense layer included)61Number of Dense Layers1Attention Hidden Dimension7168MoE Hidden Dimension(per Expert)2048Number of Attention Heads64Number of Experts384Selected Experts per Token8Number of Shared Experts1Vocabulary Size160KContext Length256KAttention MechanismMLAActivation FunctionSwiGLUVision EncoderMoonViTParameters of Vision Encoder400M

https://huggingface.co/unsloth/Kimi-K2.6-GGUF#3-evaluation-results3. Evaluation Results

BenchmarkKimi K2.6GPT-5.4 (xhigh)Claude Opus 4.6 (max effort)Gemini 3.1 Pro (thinking high)Kimi K2.5AgenticHLE-Full (w/ tools)54.052.153.051.450.2BrowseComp83.282.783.785.974.9BrowseComp (Agent Swarm)86.378.4DeepSearchQA (f1-score)92.578.691.381.989.0DeepSearchQA (accuracy)83.063.780.660.277.1WideSearch (item-f1)80.8---72.7Toolathlon50.054.647.248.827.8MCPMark55.962.5*56.7*55.9*29.5Claw Eval (pass^3)62.360.370.457.852.3Claw Eval (pass@3)80.978.482.482.975.4APEX-Agents27.933.333.032.011.5OSWorld-Verified73.175.072.7-63.3CodingTerminal-Bench 2.0 (Terminus-2)66.765.4*65.468.550.8SWE-Bench Pro58.657.753.454.250.7SWE-Bench Multilingual76.7-77.876.9*73.0SWE-Bench Verified80.2-80.880.676.8SciCode52.256.651.958.948.7OJBench (python)60.6-60.370.754.7LiveCodeBench (v6)89.6-88.891.785.0Reasoning & KnowledgeHLE-Full34.739.840.044.430.1AIME 202696.499.296.798.395.8HMMT 2026 (Feb)92.797.796.294.787.1IMO-AnswerBench86.091.475.391.0*81.8GPQA-Diamond90.592.891.394.387.6VisionMMMU-Pro79.481.273.983.0*78.5MMMU-Pro (w/ python)80.182.177.385.3*77.7CharXiv (RQ)80.482.8*69.180.2*77.5CharXiv (RQ) (w/ python)86.790.0*84.789.9*78.7MathVision87.492.0*71.2*89.8*84.2MathVision (w/ python)93.296.1*84.6*95.7*85.0BabyVision39.849.714.851.636.5BabyVision (w/ python)68.580.2*38.4*68.3*40.5V* (w/ python)96.998.4*86.4*96.9*86.9 Footnotes1. General Testing Details- 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. - 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. - 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. 2. Reasoning Benchmarks- IMO-AnswerBench scores for GPT-5.4 and Claude 4.6 were obtained fromz.ai/blog/glm-5.1. - 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. 3. Tool-Augmented / Agentic Tasks- Kimi K2.6 was equipped with search, code-interpreter, and web-browsing tools for HLE with tools, BrowseComp, DeepSearchQA, and WideSearch. - 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. - For BrowseComp, we report scores obtained with context management using the same discard-all strategy as Kimi K2.5 and DeepSeek-V3.2. - 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 theClaude Opus 4.7 System Card. - 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. - The test system prompts are identical to those used in theKimi K2.5 technical report. - Claw Eval was conducted using version 1.1 with max-tokens-per-step = 16384. - For APEX-Agents, we evaluate 452 tasks from the public 480-task release, as done byArtificial Analysis(excluding Investment Banking Worlds 244 and 246, which have external runtime dependencies) 4. Coding Tasks- Terminal-Bench 2.0 scores were obtained with the default agent framework (Terminus-2) and the provided JSON parser, operating in preserve thinking mode. - 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. - All reported scores for coding tasks are averaged over 10 independent runs. 5. Vision Benchmarks- Max-tokens = 98,304, averaged over three runs (avg@3). - Settings with Python tool use max-tokens-per-step = 65,536 and max-steps = 50 for multi-step reasoning. - MMMU-Pro follows the official protocol, preserving input order and prepending images.

https://huggingface.co/unsloth/Kimi-K2.6-GGUF#4-native-int4-quantization4. Native INT4 Quantization

Kimi-K2.6 adopts the same native int4 quantization method asKimi-K2-Thinking.

https://huggingface.co/unsloth/Kimi-K2.6-GGUF#5-deployment5. Deployment

You can access Kimi-K2.6’s API onhttps://platform.moonshot.aiand we provide OpenAI/Anthropic-compatible API for you. To verify the deployment is correct, we also provide theKimi Vendor Verifier. Currently, Kimi-K2.6 is recommended to run on the following inference engines:

• vLLM
• SGLang
• KTransformers

Kimi-K2.6 has the same architecture as Kimi-K2.5, and the deployment method can be directly reused.

The version requirement fortransformersis\>=4\.57\.1, <5\.0\.0.

Deployment examples can be found in theModel Deployment Guide.

https://huggingface.co/unsloth/Kimi-K2.6-GGUF#6-model-usage6. Model Usage

The usage demos below demonstrate how to call our official API.

For third-party APIs deployed with vLLM or SGLang, please note that:

• Chat with video content is an experimental feature and is only supported in our official API for now. - The recommendedtemperaturewill be1\.0for Thinking mode and0\.6for Instant mode. - The recommendedtop\_pis0\.95. - To use instant mode, you need to pass\{'chat\_template\_kwargs': \{"thinking": False\}\}inextra\_body.

https://huggingface.co/unsloth/Kimi-K2.6-GGUF#chat-completionChat Completion

This is a simple chat completion script which shows how to call K2.6 API in Thinking and Instant modes.

import openai
import base64
import requests
def simple_chat(client: openai.OpenAI, model_name: str):
    messages = [
        {'role': 'system', 'content': 'You are Kimi, an AI assistant created by Moonshot AI.'},
        {
            'role': 'user',
            'content': [
                {'type': 'text', 'text': 'which one is bigger, 9.11 or 9.9? think carefully.'}
            ],
        },
    ]
    response = client.chat.completions.create(
        model=model_name, messages=messages, stream=False, max_tokens=4096
    )
    print('====== Below is reasoning content in Thinking Mode ======')
    print(f'reasoning content: {response.choices[0].message.reasoning}')
    print('====== Below is response in Thinking Mode ======')
    print(f'response: {response.choices[0].message.content}')

    # To use instant mode, pass {"thinking" = {"type":"disabled"}}
    response = client.chat.completions.create(
        model=model_name,
        messages=messages,
        stream=False,
        max_tokens=4096,
        extra_body={'thinking': {'type': 'disabled'}},  # this is for official API
        # extra_body= {'chat_template_kwargs': {"thinking": False}}  # this is for vLLM/SGLang
    )
    print('====== Below is response in Instant Mode ======')
    print(f'response: {response.choices[0].message.content}')

https://huggingface.co/unsloth/Kimi-K2.6-GGUF#chat-completion-with-visual-contentChat Completion with visual content

K2.6 supports Image and Video input.

The following example demonstrates how to call K2.6 API with image input:

import openai
import base64
import requests

def chat_with_image(client: openai.OpenAI, model_name: str):
    url = 'https://huggingface.co/moonshotai/Kimi-K2.6/resolve/main/figures/kimi-logo.png'
    image_base64 = base64.b64encode(requests.get(url).content).decode()
    messages = [
        {
            'role': 'user',
            'content': [
                {'type': 'text', 'text': 'Describe this image in detail.'},
                {
                    'type': 'image_url',
                    'image_url': {'url': f'data:image/png;base64, {image_base64}'},
                },
            ],
        }
    ]

    response = client.chat.completions.create(
        model=model_name, messages=messages, stream=False, max_tokens=8192
    )
    print('====== Below is reasoning content in Thinking Mode ======')
    print(f'reasoning content: {response.choices[0].message.reasoning}')
    print('====== Below is response in Thinking Mode ======')
    print(f'response: {response.choices[0].message.content}')

    # Also support instant mode if you pass {"thinking" = {"type":"disabled"}}
    response = client.chat.completions.create(
        model=model_name,
        messages=messages,
        stream=False,
        max_tokens=4096,
        extra_body={'thinking': {'type': 'disabled'}},  # this is for official API
        # extra_body= {'chat_template_kwargs': {"thinking": False}}  # this is for vLLM/SGLang
    )
    print('====== Below is response in Instant Mode ======')
    print(f'response: {response.choices[0].message.content}')

    return response.choices[0].message.content

The following example demonstrates how to call K2.6 API with video input:

import openai
import base64
import requests

def chat_with_video(client: openai.OpenAI, model_name:str):
    url = 'https://huggingface.co/moonshotai/Kimi-K2.6/resolve/main/figures/demo_video.mp4'
    video_base64 = base64.b64encode(requests.get(url).content).decode()
    messages = [
        {
            "role": "user",
            "content": [
                {"type": "text","text": "Describe the video in detail."},
                {
                    "type": "video_url",
                    "video_url": {"url": f"data:video/mp4;base64,{video_base64}"},
                },
            ],
        }
    ]

    response = client.chat.completions.create(model=model_name, messages=messages)
    print('====== Below is reasoning content in Thinking Mode ======')
    print(f'reasoning content: {response.choices[0].message.reasoning}')
    print('====== Below is response in Thinking Mode ======')
    print(f'response: {response.choices[0].message.content}')

    # Also support instant mode if pass {"thinking" = {"type":"disabled"}}
    response = client.chat.completions.create(
        model=model_name,
        messages=messages,
        stream=False,
        max_tokens=4096,
        extra_body={'thinking': {'type': 'disabled'}},  # this is for official API
        # extra_body= {'chat_template_kwargs': {"thinking": False}}  # this is for vLLM/SGLang
    )
    print('====== Below is response in Instant Mode ======')
    print(f'response: {response.choices[0].message.content}')
    return response.choices[0].message.content

https://huggingface.co/unsloth/Kimi-K2.6-GGUF#preserve-thinkingPreserve Thinking

Kimi K2.6 supportspreserve\_thinkingmode, which retains full reasoning content across multi-turn interactions and enhances performance in coding agent scenarios.

This feature is disabled by default. The following example demonstrates how to call K2.6 API inpreserve\_thinkingmode:

def chat_with_preserve_thinking(client: openai.OpenAI, model_name: str):
    messages = [
        {
            "role": "user",
            "content": "Tell me three random numbers."
        },
        {
            "role": "assistant",
            "reasoning_content": "I'll start by listing five numbers: 473, 921, 235, 215, 222, and I'll tell you the first three.",
            "content": "473, 921, 235"
        },
        {
            "role": "user",
            "content": "What are the other two numbers you have in mind?"
        }
    ]

    response = client.chat.completions.create(
        model=model_name,
        messages=messages,
        stream=False,
        max_tokens=4096,
        extra_body={'thinking': {'type': 'enabled', 'keep': 'all'}},  # this is for official API
        # extra_body={"chat_template_kwargs": {"thinking":True, "preserve_thinking": True}},  # this is for vLLM/SGLang
        # We recommend enabling preserve_thinking only in think mode.
    )
    # the assistant should mention 215 and 222 that appear in the prior reasoning content
    print(f"response: {response.choices[0].message.reasoning}")
    return response.choices[0].message.content

https://huggingface.co/unsloth/Kimi-K2.6-GGUF#interleaved-thinking-and-multi-step-tool-callInterleaved Thinking and Multi-Step Tool Call

K2.6 shares the same design of Interleaved Thinking and Multi-Step Tool Call as K2 Thinking. For usage example, please refer to theK2 Thinking documentation.

https://huggingface.co/unsloth/Kimi-K2.6-GGUF#coding-agent-frameworkCoding Agent Framework

Kimi K2.6 works best with Kimi Code CLI as its agent framework — give it a try athttps://www.kimi.com/code.

https://huggingface.co/unsloth/Kimi-K2.6-GGUF#7-license7. License

Both the code repository and the model weights are released under theModified MIT License.

https://huggingface.co/unsloth/Kimi-K2.6-GGUF#8-third-party-notices8. Third Party Notices

SeeTHIRD PARTY NOTICES

https://huggingface.co/unsloth/Kimi-K2.6-GGUF#9-contact-us9. Contact Us

If you have any questions, please reach out at[email protected].