Variant	Generation speed	Max comfortable context	CPU offload
`Q2_K` (AllGPU)	~45 tok/s	131072	none — whole model + native MTP head resident in VRAM
`Q3_K_M` (AllGPU)	~45 tok/s	60000	none — whole model + native MTP head resident in VRAM
`Q4_K_M` (32K ctx)	~25 tok/s	32768	9 of 64 layers on CPU
`Q4_K_M` (128K ctx)	~12 tok/s	128000	30 of 64 layers on CPU

Q2 and Q3 both win by more than 4x over Q4, and it isn't really about the quant level — it's that Q4_K_M (15.66 GiB) doesn't fit a 16 GB card at all without offloading roughly half the model to CPU, while Q2_K (10.12 GiB) and Q3_K_M (12.57 GiB) fit whole. Same lesson as our MoE releases: once any layers spill to CPU, PCIe round-trips dominate and speed craters, regardless of how good the per-layer compute is. Q2's real advantage over Q3 isn't raw speed at the same context — it's that its smaller footprint buys more than double the usable context (131072 vs 60000) before you'd need to fall back to CPU offload, at comparable throughput.

Tuning `--spec-draft-n-max` per file

We swept the MTP draft length on each file (10K-token prompt, same hardware) instead of assuming one value fits all — the optimum turned out to be file-dependent, not a constant:

Q2_K (60000 ctx, AllGPU):

`--spec-draft-n-max`	Speed	Draft acceptance
1	31.5 tok/s	108/134 (80.6%)
2	34.4 tok/s	134/227 (59.0%)
3	35.4 tok/s (peak)	143/313 (45.7%)
4	30.6 tok/s	142/418 (34.0%)
6	23.2 tok/s	145/604 (24.0%)

Q3_K_M (60000 ctx, AllGPU):

`--spec-draft-n-max`	Speed	Draft acceptance
1	28.8 tok/s	111/137 (81.0%)
2	37.7 tok/s (peak)	136/225 (60.4%)
3	31.2 tok/s	148/300 (49.3%)
4	32.9 tok/s	153/383 (39.9%)
6	crashed (CUDA OOM mid-generation)	—

Q4_K_M (128000 ctx, 30 layers on CPU):

`--spec-draft-n-max`	Speed	Draft acceptance
1	9.2 tok/s	110/139 (79.1%)
2	12.2 tok/s (peak — also the VRAM ceiling)	143/211 (67.8%)
3	crashed (CUDA OOM mid-generation, ~19% into an 8192-token prompt-processing batch)	—

Q2_K tolerates the deepest draft (n-max 3) because its coarser quantization already accepts fewer drafts per step, so there's more room before the curve turns over. Q3_K_M and Q4_K_M both peak at n-max 2 — for Q3_K_M this is a genuine speed optimum (n-max 3 is measurably slower, not just untested); for Q4_K_M it's also a hard VRAM ceiling — the 30-layers-on-GPU offload split has no headroom left for a deeper draft context, so n-max 3 doesn't run at all. In every case, acceptance rate falls faster than draft length grows once you're past the peak, so verification cost outpaces tokens actually saved — don't assume a bigger n-max is free speed.

Tested hardware and software

Linux
NVIDIA GeForce RTX 5060 Ti, 16 GB VRAM
AMD Ryzen 9 5950X, 16 cores / 32 threads
64 GB system RAM
TheTom/llama-cpp-turboquant, commit c26cbdffc

This is a dense model (no MoE experts), so there's no _exps/--n-cpu-moe machinery here — CPU offload, where used, targets plain ffn_gate/ffn_up/ffn_down tensors by layer index instead.

About this release

Qwen3.8-27B ships upstream as a vision-language checkpoint (Qwen3_5ForConditionalGeneration, ~333 vision-tower tensors). These GGUF files are text-only: the converter extracts just text_config and never touches the vision tower, so nothing needed stripping after the fact — the vision tensors simply never entered the pipeline. Verified directly against the converted file: 0 vision/image/video tensors present.

Unlike our Ornith/KAT-Coder MTP releases, no head-grafting was needed here. Qwen3.8-27B ships its own natively-trained Multi-Token Prediction layer (mtp_num_hidden_layers: 1 in the upstream config) as the model's own extra transformer block (blk.64). It quantizes and travels through this pipeline like any other layer — every file below already has real, natively-trained speculative decoding built in via --spec-type draft-mtp.

Files

File	Size	BPW	Recipe
`Qwen3.8-27B_Q4_K_M.gguf`	15.66 GiB	4.92	Plain `Q4_K_M` + iMatrix, no manual overrides
`Qwen3.8-27B_Q3_K_M.gguf`	12.57 GiB	3.95	Plain `Q3_K_M` + iMatrix, no manual overrides
`Qwen3.8-27B_Q2_K.gguf`	10.12 GiB	3.18	Plain `Q2_K` + iMatrix, no manual overrides

All three calibrated with the same iMatrix pipeline as our other releases (calibration_datav5.txt, 802 chunks). No per-tensor overrides on any file — dense models don't need the routed-expert-specific tuning our MoE releases use, the stock K-quant preset's own position-dependent mix (protecting early/late layers) already does sensible work here.

Recommended launch commands

Q2_K — everything on GPU, largest comfortable context

bash

llama-server \
  --jinja --host 0.0.0.0 --port 8080 \
  -m Qwen3.8-27B_Q2_K.gguf \
  --spec-type draft-mtp --spec-draft-n-max 3 \
  --n-gpu-layers 99 \
  --ctx-size 131072 --parallel 1 \
  --flash-attn on \
  --cache-type-k turbo3 --cache-type-v turbo3 \
  --batch-size 262144 --ubatch-size 1024 --cache-reuse 256

--spec-draft-n-max 3 is the measured optimum (see sweep above). No -ot line — whole model plus native MTP head fit in VRAM at this context. This is the file to reach for if you want more headroom than Q3's 60000-token ceiling without paying Q4's CPU-offload tax.

Q3_K_M — everything on GPU (fastest at moderate context)

bash

llama-server \
  --jinja --host 0.0.0.0 --port 8080 \
  -m Qwen3.8-27B_Q3_K_M.gguf \
  --spec-type draft-mtp --spec-draft-n-max 2 \
  --n-gpu-layers 99 \
  --ctx-size 60000 --parallel 1 \
  --flash-attn on \
  --cache-type-k turbo3 --cache-type-v turbo3 \
  --batch-size 60000 --ubatch-size 1024 --cache-reuse 256

No -ot line — the whole model plus its native MTP head fit in VRAM outright. Context is capped at 60000 (well under the model's 262144 native max) to leave headroom for the KV cache and the MTP draft context's own cache, both of which scale with context length on this architecture.

Q4_K_M — 32K context, lighter offload (faster)

bash

llama-server \
  --jinja --host 0.0.0.0 --port 8080 \
  -m /home/maxim/models/gguf/Qwen3.8-27B_Q4_K_M_imatrix.gguf \
  --spec-type draft-mtp --spec-draft-n-max 2 \
  --n-gpu-layers 99 \
  -ot "blk\.(5[5-9]|6[0-3])\.ffn_(gate|up|down)\.weight=CPU" \
  --ctx-size 32768 --parallel 1 \
  --flash-attn on \
  --cache-type-k turbo3 --cache-type-v turbo3 \
  --batch-size 8192 --ubatch-size 512 --cache-reuse 256

Only 9 of 64 layers (blk.55-blk.63) offloaded instead of 30 — the smaller 32K context needs far less KV-cache VRAM, leaving room to keep most of the model on GPU. Measured ~25 tok/s at 10K context — roughly 2x the 128K profile's throughput. Reach for this over the 128K profile whenever 32K context is enough for the task.

Q4_K_M — 128K context, CPU offload

bash

llama-server \
  --jinja --host 0.0.0.0 --port 8080 \
  -m Qwen3.8-27B_Q4_K_M.gguf \
  --spec-type draft-mtp --spec-draft-n-max 2 \
  --n-gpu-layers 99 \
  -ot "blk\.(3[4-9]|4[0-9]|5[0-9]|6[0-3])\.ffn_(gate|up|down)\.weight=CPU" \
  --ctx-size 128000 --parallel 1 \
  --flash-attn on \
  --cache-type-k turbo3 --cache-type-v turbo3 \
  --batch-size 8192 --ubatch-size 1024 --cache-reuse 256

30 of 64 layers offloaded to CPU RAM to fit the larger context — --batch-size 8192 (not the 262144 used on the AllGPU files) matters here: an oversized batch relative to actual usage wastes VRAM on compute buffers that this offload split doesn't have to spare, and n-max 3 will not load at all at this offload level (confirmed — see the sweep above). Only reach for this file over Q3 if you specifically need the extra quality headroom and can tolerate roughly a 3x slowdown.

Upstream model card

This release is based on the original Qwen3.8-27B model card reproduced in full below. Its original license declaration and complete model card are preserved unchanged.

Qwen3.8-27B

[!Note] This repository contains model weights and configuration files for the post-trained model in the Hugging Face Transformers format.

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

[!Tip] For users seeking managed, scalable inference without infrastructure maintenance, the official Qwen API service is provided by Qwen Cloud. In particular, Qwen3.8-27B will be available as a hosted version with more production features, e.g., 1M context length by default, official built-in tools. For more information, please refer to the Qwen3.8-27B Overview. The service is coming soon. Stay tuned for updates.

Following the widespread community adoption of the Qwen3.5 and Qwen3.6 series, we are pleased to introduce Qwen3.8, the most capable generation in the Qwen open-model family to date.

Built on the architectural foundation of Qwen3.5, Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks. Qwen3.8-27B brings these advances to a compact, deployment-friendly dense model: a native vision-language model that understands images and videos, with flexible thinking control, designed to carry complex, multi-step tasks through to completion with greater reliability.

Qwen3.8 Highlights

Qwen3.8-27B features the following enhancements:

Core Capabilities: Comprehensive improvements across coding, professional work, research, and long-horizon agentic tasks.
Agent Execution: Stronger autonomous planning and better handling of environment feedback, leading to more reliable end-to-end task completion.
Downstream Compatibility: Broader support for popular harnesses and development tools, making it easier to integrate into your existing stack.
Flexible Thinking Control: Thinking mode is on by default and can be disabled per request; reasoning depth can be tuned with reasoning_effort, and reasoning context from historical messages is retained via preserve_thinking.
Vision-Language Understanding: Native support for image and video understanding, from STEM diagrams and documents to hour-scale videos.

Model Overview

Type: Causal Language Model with Vision Encoder
Training Stage: Pre-training & Post-training
Language Model
- Number of Parameters: 27B
- Hidden Dimension: 5120
- Token Embedding: 248,320 (Padded)
- Number of Layers: 64
- Hidden Layout: 16 × (3 × (Gated DeltaNet → FFN) → 1 × (Gated Attention → FFN))
- Gated DeltaNet:
  - Number of Linear Attention Heads: 48 for V and 16 for QK
  - Head Dimension: 128
- Gated Attention:
  - Number of Attention Heads: 24 for Q and 4 for KV
  - Head Dimension: 256
  - Rotary Position Embedding Dimension: 64
- Feed Forward Network:
  - Intermediate Dimension: 17,408
- LM Output: 248,320 (Padded)
- MTP (Multi-Token Prediction): trained with multiple steps
Context Length: 262,144 natively and extensible up to 1,000,000 tokens.

Benchmark Results

Text Performance

	Qwen3.8-27B	Qwen3.6-27B	Qwen3.7-Plus	Muse Glimmer-30B	Opus4.6 Max
Coding
Agentic terminal coding Terminal Bench 2.1 (Terminus)	73.0	63.4	64.0	51.7	78.2
Agentic coding SWE-bench Pro	61.7	53.5	57.6	51.2	53.4
Repo-level code generation NL2Repo-Bench	42.3	36.2	41.1	--	47.6
Agentic coding DeepSWE 1.1	42.2	13.3	14.2	--	--
Software engineering QwenSWEBench	79.0	49.3	59.2	--	63.8
Agent
Long-horizon office work CoWorkBench	70.7	61.0	65.1	--	68.2
Professional job tasks JobBench	33.4	21.8	27.6	--	--
Frontier agentic tasks Agents' Last Exam	Pass@1 20.4 Score 42.9	Pass@1 10.6 Score 27.3	Pass@1 13.2 Score 33.6	--	--
General
Instruction following IFBench	79.5	69.1	79.1	77.0	62.5
Scientific reasoning GPQA Diamond	89.2	87.8	90.3	83.5	91.3
Multidisciplinary reasoning HLE	30.8	24.0	34.7	22.0	40.0
Competitive coding LiveCodeBench v6	90.3	83.9	89.6	--	88.8

SWE-bench Pro: Except for Opus4.6 Max, which uses the officially reported score, all models are evaluated with the Claude Code harness at temp=1.0, top_p=0.95, and a 256K context window. Problematic tasks were corrected, and all baseline models were re-evaluated on the refined benchmark.
NL2Repo-Bench: Evaluated with the Claude Code harness. To prevent reward hacking, we disable Bash commands that attempt to access the specific repository, such as pip download, pip install, and git clone.
DeepSWE 1.1: Evaluated with the Claude Code harness at temp=1.0, top_p=0.95, and a 256K context window.
QwenSWEBench: In-house coding benchmark for evaluating models' software engineering capabilities. Evaluated with the Claude Code harness. Reporting avg@3 with an 8-hour timeout, max_tokens=32,768, temperature=1.0, and a 256K context window.
CoWorkBench: In-house cowork benchmark for evaluating long-horizon tasks across computer science, finance, law, medical, and other productivity domains.
HLE: Judged by GPT-4o.
The best result in each row is shown in bold.
Empty cells (--) indicate that results are not yet available or not applicable.

VL Performance

	Qwen3.8-27B	Qwen3.6-27B	Qwen3.7-Plus	Muse Glimmer-30B	Opus4.6 Max
Agentic Multimodal Intelligence
Computer use OSWorld-Verified	84.3	63.9	73.3	65.9	72.7
Browser use WebArena-Verified	64.8	48.8	55.3	--	--
Mobile use AndroidWorld	81.9	70.3	81.0	--	62.0
Application recreation RecreationBench	47.1	29.8	30.2	--	--
Multimodal tool use ClawEval-MM	Pass@3 57.4 Average 56.9	Pass@3 42.6 Average 50.4	Pass@3 57.4 Average 60.1	--	Pass@3 52.5 Average 54.7
Multimodal software engineering SWE-MM	38.6	25.7	30.0	--	27.1
Visual web development Vision2Web	62.9	45.0	42.1	--	--
General Multimodal Intelligence
Visual math problem solving MathVision	Without CI 90.0 With CI 94.6	Without CI 85.1	Without CI 90.3	--	Without CI 65.5
General visual reasoning BabyVision	Without CI 65.7 With CI 85.6	Without CI 28.9	Without CI 64.7 With CI 70.4	--	Without CI 12.6
Scientific chart analysis CharXiv (RQ)	Without CI 83.7 With CI 90.2	Without CI 78.4	Without CI 85.8 With CI 85.9	78.8	Without CI 66.0
Document intelligence OmniDocBench 1.5	91.1	89.4	91.4	75.8	86.6
Real-world perception RealWorldQA	85.9	84.1	86.9	--	73.9
Embodied intelligence ERQA	65.5	62.5	69.8	--	40.8

MathVision, BabyVision, and CharXiv (RQ): Where both settings are available, cells report “Without CI” and “With CI” separately; otherwise, only the available setting is shown. A small number of incorrect ground-truth annotations in MathVision and CharXiv (RQ) were corrected following manual verification, and all reported scores on those benchmarks were computed using the corrected annotations.
MathVision: Qwen3.8-27B is evaluated using the fixed prompt: “Please reason step by step, and put your final answer within \boxed{}.” For the remaining models, we report the higher score from two prompt variants—one with and one without the \boxed{} formatting requirement.
WebArena-Verified: Scores are computed with the official WebArena-Verified grader under the OSWorld scaffold.
RecreationBench: An in-house, long-horizon application-recreation benchmark designed to evaluate hybrid-agent capabilities across five platforms: desktop (Ubuntu, macOS, and Windows), mobile (Android), and the web.
ClawEval-MM: Scores are reported as “Pass@3 / average score.” Pass@3 is the percentage of tasks passed in at least one of three trials; the average score is the mean benchmark score across the three trials.
Vision2Web: Scores are averaged across the frontend, webpage, and website categories. Evaluations use the Claude Code harness and are judged by gpt-5.4-2026-03-05.
SWE-MM: Scores are evaluated on the Claude Code harness using the public dev split of SWE-bench Multimodal, with the modifications described in Appendix 8.3 of the Claude Opus 4.7 system card.
Empty cells (--) indicate that results are not yet available or not applicable.

Quickstart

For streamlined integration, we recommend using Qwen3.8 via APIs.

Serving Qwen3.8

[!Important] Inference efficiency and throughput vary significantly across frameworks. We recommend using the latest framework versions to ensure optimal performance and compatibility. For production workloads or high-throughput scenarios, dedicated serving engines such as SGLang, vLLM, or TokenSpeed are recommended.

Qwen3.8 can be deployed with popular inference frameworks, e.g.:

API Usage

[!Important] Qwen3.8 models operate in thinking mode by default, generating thinking content signified by <think>\n...</think>\n\n before producing the final response. To disable thinking content and obtain a direct response, refer to the examples here.

[!Tip] We recommend using the following sets of sampling parameters for generation:

Thinking Mode: temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0

Instruct (or non-thinking) mode: temperature=0.7, top_p=0.80, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0

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

Qwen3.8 comes with official support for reasoning_effort, which can be used to adjust reasoning depth and control cost:

xhigh (default): for complex tasks demanding thorough analysis
medium: balancing accuracy and speed
low: efficient reasoning optimizing for speed and cost

In addition, preserve_thinking is enabled by default for all workloads for the best out-of-the-box experience. To disable preserved thinking, refer to the examples here.

[!Tip] In multi-turn agentic tasks, lower reasoning effort does not always reduce overall task completion time. Although it may produce faster per-turn responses, it can also lead to insufficient analysis, more failures, and repeated retries, which may increase total latency and token consumption.

Chat Completions API

The Chat Completions API can be used with most inference frameworks, as well as Qwen Cloud. Before starting, make sure the OpenAI Python SDK is installed and the API key and the API base URL are configured, e.g.:

shell

pip install -U openai

# Set the following accordingly
export OPENAI_BASE_URL='your-base-url'
export OPENAI_API_KEY='your-api-key'

Text-Only Input

python

from openai import OpenAI
# Configured by environment variables
client = OpenAI()

messages = [{"role": "user", "content": "Write a Python function to merge two sorted linked lists."}]

completion = client.chat.completions.create(
    model="Qwen/Qwen3.8-27B",
    messages=messages,
    extra_body={
        "chat_template_kwargs": {
            "enable_thinking": True,  # on by default
            "preserve_thinking": True, # on by default
        },
    },
    reasoning_effort="xhigh",  # xhigh by default; supported levels are xhigh, medium, and low
    stream=True,
    stream_options={"include_usage": True},
)

reasoning_content = ""
answer_content = ""
is_answering = False
print("\n" + "=" * 20 + "Reasoning" + "=" * 20 + "\n")

for chunk in completion:
    if not chunk.choices:
        print("\nUsage:")
        print(chunk.usage)
        continue

    delta = chunk.choices[0].delta

    if hasattr(delta, "reasoning_content") and delta.reasoning_content is not None:
        if not is_answering:
            print(delta.reasoning_content, end="", flush=True)
        reasoning_content += delta.reasoning_content
    elif hasattr(delta, "reasoning") and delta.reasoning is not None:
        if not is_answering:
            print(delta.reasoning, end="", flush=True)
        reasoning_content += delta.reasoning

    if hasattr(delta, "content") and delta.content:
        if not is_answering:
            print("\n" + "=" * 20 + "Answer" + "=" * 20 + "\n")
            is_answering = True
        print(delta.content, end="", flush=True)
        answer_content += delta.content

messages.append({
    "role": "assistant",
    "content": answer_content,
    "reasoning_content": reasoning_content,
    "reasoning": reasoning_content,
})

Image Input

python

from openai import OpenAI
# Configured by environment variables
client = OpenAI()

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image_url",
                "image_url": {
                    "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/CI_Demo/mathv-1327.jpg"
                }
            },
            {
                "type": "text",
                "text": "The centres of the four illustrated circles are in the corners of the square. The two big circles touch each other and also the two little circles. With which factor do you have to multiply the radii of the little circles to obtain the radius of the big circles?\nChoices:\n(A) $\\frac{2}{9}$\n(B) $\\sqrt{5}$\n(C) $0.8 \\cdot \\pi$\n(D) 2.5\n(E) $1+\\sqrt{2}$"
            }
        ]
    }
]

chat_response = client.chat.completions.create(
    model="Qwen/Qwen3.8-27B",
    messages=messages,
)
print("Chat response:", chat_response)

Video Input

python

from openai import OpenAI
# Configured by environment variables
client = OpenAI()

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "video_url",
                "video_url": {
                    "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/video/N1cdUjctpG8.mp4"
                }
            },
            {
                "type": "text",
                "text": "How many porcelain jars were discovered in the niches located in the primary chamber of the tomb?"
            }
        ]
    }
]

chat_response = client.chat.completions.create(
    model="Qwen/Qwen3.8-27B",
    messages=messages,
)

# When vLLM is launched with `--media-io-kwargs '{"video": {"num_frames": -1}}'`,
# video frame sampling can be configured via `extra_body` (e.g., by setting `fps`).
# This feature is currently supported only in vLLM.
#
# By default, `fps=2` and `do_sample_frames=True`.
# With `do_sample_frames=True`, you can customize the `fps` value to set your desired video sampling rate.
# chat_response = client.chat.completions.create(
#     model="Qwen/Qwen3.8-27B",
#     messages=messages,
#     extra_body={
#         "mm_processor_kwargs": {"fps": 2, "do_sample_frames": True},
#     }, 
# )

print("Chat response:", chat_response)

Instruct (or Non-Thinking) Mode

Qwen3.8-27B will think by default before responding. You can obtain a direct response from the model without thinking by configuring the API parameters. For example,

python

from openai import OpenAI
# Configured by environment variables
client = OpenAI()

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image_url",
                "image_url": {
                    "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/RealWorld/RealWorld-04.png"
                }
            },
            {
                "type": "text",
                "text": "Where is this?"
            }
        ]
    }
]

chat_response = client.chat.completions.create(
    model="Qwen/Qwen3.8-27B",
    messages=messages,
    temperature=0.7,
    top_p=0.8,
    presence_penalty=1.5,
    extra_body={
        "top_k": 20,
        "chat_template_kwargs": {"enable_thinking": False},
    }, 
)
print("Chat response:", chat_response)

[!Note] If you are using APIs from Qwen Cloud, in addition to changing model, please use "enable_thinking": False instead of "chat_template_kwargs": {"enable_thinking": False}.

Disable Preserved Thinking

By default, Qwen3.8 retains thinking blocks from all historical messages, maintaining a complete reasoning trace across the conversation. This behavior, known as preserved thinking, ensures full context continuity and is especially beneficial for agent scenarios where decision consistency and reduced redundant reasoning are critical. It also improves KV cache utilization, optimizing inference efficiency in both thinking and non-thinking modes.

If you prefer to retain only the thinking blocks from the latest user message, you can disable this behavior by setting preserve_thinking to False:

python

from openai import OpenAI

# Configured by environment variables
client = OpenAI()
messages = [...]
chat_response = client.chat.completions.create(
    model="Qwen/Qwen3.8-27B",
    messages=messages,
    extra_body={
        "chat_template_kwargs": {"preserve_thinking": False},
    },
)
print("Chat response:", chat_response)

[!Note] If you are using APIs from Qwen Cloud, in addition to changing model, please use "preserve_thinking": False directly instead of wrapping it in chat_template_kwargs.

Best Practices

To achieve optimal performance, we recommend the following settings:

Sampling Parameters: We suggest using the following sets of sampling parameters:
- Thinking Mode: temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0
- Instruct (or non-thinking) mode: temperature=0.7, top_p=0.80, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
For supported frameworks, you can adjust the presence_penalty parameter between 0 and 2 to reduce endless repetition. However, using a higher value may occasionally result in language mixing and a slight decrease in model performance.
Adequate Output Length: To optimize performance on agentic tasks, we recommend allocating sufficient output length to allow the model to generate detailed and comprehensive responses. For frameworks that support separate token limits for internal reasoning and final outputs, we suggest the following configuration within the 1M context length:
- Reasoning Content: Set the maximum output length to 262,144 tokens.
- Final Response: Set the maximum output length to 131,072 tokens.
These settings provide the necessary capacity for complex reasoning while ensuring ample space for high-quality final deliverables.
Processing Ultra-Long Texts: Qwen3.8-27B natively supports context lengths of up to 262,144 tokens. For long-horizon tasks where the total length (including both input and output) exceeds this limit, we recommend using RoPE scaling techniques to handle long texts effectively, e.g., YaRN.

YaRN is currently supported by several inference frameworks, e.g., vLLM, SGLang, and TokenSpeed. In general, there are two approaches to enabling YaRN for supported frameworks:
- Modifying the model configuration file:
  
  In the config.json file, change the rope_parameters fields in text_config to:
  
  json
  1{ 2 "mrope_interleaved": true, 3 "mrope_section": [ 4 11, 5 11, 6 10 7 ], 8 "rope_type": "yarn", 9 "rope_theta": 10000000, 10 "partial_rotary_factor": 0.25, 11 "factor": 4.0, 12 "original_max_position_embeddings": 262144, 13}
- Passing command line arguments:
  
  For vLLM, you can use
  
  VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 vllm serve ... --hf-overrides '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --max-model-len 1000000
  
  For SGLang, you can use
  
  SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 python -m sglang.launch_server ... --json-model-override-args '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --context-length 1000000
  
  For TokenSpeed, you can use
  
  TOKENSPEED_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 tokenspeed serve ... --hf-overrides '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --max-model-len 1000000
[!NOTE] All the notable open-source frameworks implement static YaRN, which means the scaling factor remains constant regardless of input length, potentially impacting performance on shorter texts. We advise modifying the rope_parameters configuration only when processing long contexts is required. It is also recommended to modify the factor as needed. For example, if the typical context length for your application is 524,288 tokens, it would be better to set factor as 2.0.
Long Video Understanding: To optimize inference efficiency for plain text and images, the size parameter in the released video_preprocessor_config.json is conservatively configured. It is recommended to set the longest_edge parameter in the video_preprocessor_config file to 469,762,048 (corresponding to 224k video tokens) to enable higher frame-rate sampling for hour-scale videos and thereby achieve superior performance. For example,

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

Alternatively, override the default values via engine startup parameters. For implementation details, refer to: vLLM / SGLang.

Citation

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

bibtex

@misc{qwen38,
    title = {{Qwen3.8-Max}: A New Bar for Coding and Cowork},
    url = {https://qwen.ai/blog?id=qwen3.8},
    author = {{Qwen Team}},
    month = {August},
    year = {2026}
}

offmonreal/Qwen3.8-27B-MaxQuality-iMatrix-GGUF

Qwen3.8-27B MaxQuality iMatrix GGUF

⚡ Speed at a glance (10K context)

Tuning --spec-draft-n-max per file

Tested hardware and software

About this release

Files

Recommended launch commands

Q2_K — everything on GPU, largest comfortable context

Q3_K_M — everything on GPU (fastest at moderate context)

Q4_K_M — 32K context, lighter offload (faster)

Q4_K_M — 128K context, CPU offload

Upstream model card

Qwen3.8-27B

Qwen3.8 Highlights

Model Overview

Benchmark Results

Text Performance

VL Performance

Quickstart

Serving Qwen3.8

API Usage

Chat Completions API

Text-Only Input

Image Input

Video Input

Instruct (or Non-Thinking) Mode

Disable Preserved Thinking

Best Practices

Citation

offmonreal/
Qwen3.8-27B-MaxQuality-iMatrix-GGUF

Tuning `--spec-draft-n-max` per file