What is unsloth/Qwen3.8-2.4T-A95B-GGUF?

--- library_name: transformers license: other license_name: qwen3.8-max license_link: LICENSE pipeline_tag: text-generation base_model: - Qwen/Qwen3.8-2.4T-A95B tags: - unsloth --- # Read our How to [Run Qwen3.8 Guide!](https://unsloth.ai/docs/models/qwen3.8) Unsloth Dynamic 2.0 achieves superior accuracy & outperforms other leading quants. <img src="https://r…

What license applies to unsloth/Qwen3.8-2.4T-A95B-GGUF?

License: other. Verify terms on Hugging Face before commercial use.

How do I run unsloth/Qwen3.8-2.4T-A95B-GGUF locally?

Download a GGUF file from this page and load it in guIDE or llama.cpp. Pipeline task: text-generation.

Model Intelligence Sheet

unsloth/Qwen3.8-2.4T-A95B-GGUF overview

Q: What license applies to unsloth/Qwen3.8-2.4T-A95B-GGUF?

License: other. Verify terms on Hugging Face before commercial use.

Q: How do I run unsloth/Qwen3.8-2.4T-A95B-GGUF locally?

Download a GGUF file from this page and load it in guIDE or llama.cpp. Pipeline task: text-generation.

Q: How much VRAM or disk space does unsloth/Qwen3.8-2.4T-A95B-GGUF need?

Runs locally from ~10.4 MB disk (4 GB VRAM class GPUs with llama.cpp / guIDE).

Read our How to Run Qwen3.8 Guide https://unsloth.ai/docs/models/qwen3.8 <p style="margin top: 0;margin bottom: 0;" <em <a href="https://unsloth.ai/docs/basics…

transformersggufunslothtext-generationbase_model:Qwen/Qwen3.8-2.4T-A95Bbase_model:quantized:Qwen/Qwen3.8-2.4T-A95Blicense:otherendpoints_compatibleregion:usconversational

Runs locally from ~10.4 MB disk (4 GB VRAM class GPUs with llama.cpp / guIDE).

Downloads

Likes

Pipeline

text-generation

Author

unsloth

Repository Files & Downloads

120 GGUF files detected

Direct downloads for local inference

File	Type	Quantization	Size	Link
BF16/Qwen3.8-2.4T-A95B-BF16-00001-of-00140.gguf	GGUF	BF16	32.01 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00002-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00003-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00004-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00005-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00006-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00007-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00008-of-00140.gguf	GGUF	BF16	35.75 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00009-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00010-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00011-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00012-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00013-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00014-of-00140.gguf	GGUF	BF16	35.73 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00015-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00016-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00017-of-00140.gguf	GGUF	BF16	38.42 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00018-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00019-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00020-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00021-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00022-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00023-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00024-of-00140.gguf	GGUF	BF16	35.75 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00025-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00026-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00027-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00028-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00029-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00030-of-00140.gguf	GGUF	BF16	35.70 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00031-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00032-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00033-of-00140.gguf	GGUF	BF16	34.63 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00034-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00035-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00036-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00037-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00038-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00039-of-00140.gguf	GGUF	BF16	35.67 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00040-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00041-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00042-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00043-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00044-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00045-of-00140.gguf	GGUF	BF16	35.67 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00046-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00047-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00048-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00049-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00050-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00051-of-00140.gguf	GGUF	BF16	35.67 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00052-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00053-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00054-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00055-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00056-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00057-of-00140.gguf	GGUF	BF16	35.67 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00058-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00059-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00060-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00061-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00062-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00063-of-00140.gguf	GGUF	BF16	35.67 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00064-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00065-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00066-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00067-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00068-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00069-of-00140.gguf	GGUF	BF16	35.67 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00070-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00071-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00072-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00073-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00074-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00075-of-00140.gguf	GGUF	BF16	35.67 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00076-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00077-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00078-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00079-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00080-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00081-of-00140.gguf	GGUF	BF16	35.67 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00082-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00083-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00084-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00085-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00086-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00087-of-00140.gguf	GGUF	BF16	35.67 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00088-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00089-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00090-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00091-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00092-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00093-of-00140.gguf	GGUF	BF16	35.67 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00094-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00095-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00096-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00097-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00098-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00099-of-00140.gguf	GGUF	BF16	35.67 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00100-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00101-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00102-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00103-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00104-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00105-of-00140.gguf	GGUF	BF16	35.67 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00106-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00107-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00108-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00109-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00110-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00111-of-00140.gguf	GGUF	BF16	35.67 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00112-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00113-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00114-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00115-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00116-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00117-of-00140.gguf	GGUF	BF16	35.67 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00118-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00119-of-00140.gguf	GGUF	BF16	32.00 GB	Download
BF16/Qwen3.8-2.4T-A95B-BF16-00120-of-00140.gguf	GGUF	BF16	32.00 GB	Download

Model Details

Model ID	unsloth/Qwen3.8-2.4T-A95B-GGUF
Author	unsloth
Pipeline	text-generation
License	other
Base model	Qwen/Qwen3.8-2.4T-A95B
Last modified	2026-08-13T01:01:45.000Z

Model README

---

library_name: transformers

license: other

license_name: qwen3.8-max

license_link: LICENSE

pipeline_tag: text-generation

base_model:

Qwen/Qwen3.8-2.4T-A95B

tags:

unsloth

---

Read our How to Run Qwen3.8 Guide!

<em><a href="https://unsloth.ai/docs/basics/unsloth-dynamic-v2.0-gguf">Unsloth Dynamic 2.0</a> achieves superior accuracy & outperforms other leading quants.</em>

</p>

</a>

</a>

</a>

</div>

<li>See our <a href="https://unsloth.ai/docs/models/qwen3.8">Qwen3.8 guide</a> for quantization analysis and instructions.</li>

<li>You can now run Qwen3.8 in <a href="https://github.com/unslothai/unsloth/">Unsloth Desktop</a> with toggles for thinking.</li>

<li>New sub IQ1_S data-types Q1_0 <a href="https://unsloth.ai/docs/models/qwen3.8#new-sub-1-bit-data-types">(IQ1_XXXXS)</a> for Qwen 3.8</li>

<li>See below for 1-bit Qwen3.8 run inside of Unsloth:</li>

</div>

---

Qwen3.8-2.4T-A95B

![Qwen Studio](https://chat.qwen.ai/?models=qwen3.8-max)

> [!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 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-Max is the official version based on Qwen3.8-2.4T-A95B with more features, such as vision input & non-thinking support, 1M context length by default, official built-in tools, etc.

> For more information, please refer to the Qwen3.8-Max Overview.

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.

For the first time, Qwen3.8 brings a Qwen-Max-class model to open release.

Built on the architectural foundation of Qwen3.5, Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks. Beyond answering harder questions, Qwen3.8 is designed to carry complex, multi-step tasks through to completion with greater reliability.

Qwen3.8 Highlights

Qwen3.8 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: Reasoning depth can be tuned with reasoning_effort, and reasoning context from historical messages is retained via preserve_thinking.

For more details, please refer to our blog post Qwen3.8-Max.

Model Overview

Type: Causal Language Model
Training Stage: Pre-training & Post-training
Language Model

- Number of Parameters: 2.4T in total and 95B activated

- Hidden Dimension: 8192

- Token Embedding: 248,320 (Padded)

- Number of Layers: 92

- Hidden Layout: 23 × (3 × (Gated DeltaNet → MoE) → 1 × (Gated Attention → MoE))

- Gated DeltaNet:

- Number of Linear Attention Heads: 128 for V and 16 for QK

- Head Dimension: 128

- Gated Attention:

- Number of Attention Heads: 64 for Q and 4 for KV

- Head Dimension: 256

- Rotary Position Embedding Dimension: 64

- Mixture of Experts:

- Number of Experts: 512

- Number of Activated Experts: 10 Routed + 1 Shared

- Expert Intermediate Dimension: 2048

- LM Output: 248,320 (Padded)

- MTP (Multi-Token Prediction): trained with multiple steps

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

Benchmark Results

<th style="padding:10px 7px;text-align:left;font-weight:600;border-bottom:2px solid #0A2EFE;color:#0A2EFE"></th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #0A2EFE;color:#0A2EFE;font-size: 14px;width:14.00%;">Opus 4.8</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #0A2EFE;color:#0A2EFE;font-size: 14px;width:14.00%;">Fable 5</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #0A2EFE;color:#0A2EFE;font-size: 14px;width:14.00%;">GPT 5.6 Sol (max)</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #0A2EFE;color:#0A2EFE;font-size: 14px;width:14.00%;">Qwen3.7-Max</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #0A2EFE;color:#0A2EFE;font-size: 14px;width:14.00%;background:rgba(10, 46, 254, 0.08);">Qwen3.8-Max</th></tr></thead>

<tbody>

<tr><td colspan="6" style="padding:8px 12px;font-weight:600;color:#0A2EFE;border-bottom:1px solid rgba(10, 46, 254, 0.2);background:#D6DAFC">Coding Agent</td></tr>

<tr>

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Terminal Bench 2.1</td>

</tr>

<tr>

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

</tr>

<tr>

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">DeepSWE 1.1</td>

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr><td colspan="6" style="padding:8px 12px;font-weight:600;color:#0A2EFE;border-bottom:1px solid rgba(10, 46, 254, 0.2);background:#D6DAFC">General Agent</td></tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Agents' Last Exam (Pass / Score)</td>

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr><td colspan="6" style="padding:8px 12px;font-weight:600;color:#0A2EFE;border-bottom:1px solid rgba(10, 46, 254, 0.2);background:#D6DAFC">General Capabilities</td></tr>

<tr>

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

</tr>

<tr>

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

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

</tr>

<tr>

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MRCR v2 256K (8-needle)</td>

</tr>

<tr>

<td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">LongBench v2</td>

</tr>

</tbody>

</table>

<p style="margin-top:12px;font-size:10px;line-height:1.4;opacity:0.7">1. Fable5 results may involve fallbacks.<br>2. Terminal Bench 2.1: Evaluated with Claude Code (avg@10), using a 5-hour timeout and max_tokens=131,072. For all other models, we report the best published score across harnesses: Claude Opus 4.8 and Claude Fable 5 with Terminus 2 from Artificial Analysis (https://artificialanalysis.ai/evaluations/terminalbench-v2-1); GPT-5.6 Sol with Codex (https://openai.com/index/previewing-gpt-5-6-sol/).<br>3. SWE-bench Pro: Evaluated with the Claude Code harness, temp=1.0, top_p=0.95, and a 256K context window. Problematic tasks corrected and all baselines evaluated on the refined benchmark.<br>4. DeepSWE 1.1: Evaluated with the Claude Code and mini-SWE-agent harnesses, temp=1.0, top_p=0.95, and a 256K context window. We report the highest score among both harnesses; notably, Qwen3.8-Max performs best on Claude Code.<br>5. 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.<br>6. FrontierSWE: Evaluated with the Claude Code harness. All other available MEAN@5 results are taken from the official FrontierSWE leaderboard (https://www.frontierswe.com) as of August 3, 2026. Dominance scores are recomputed from the raw scores using the official evaluation script. "--" indicates that no official MEAN@5 result was available as of that date.<br>7. MLS-Bench-Lite: Evaluated with Claude Code using a 5-hour timeout and max_tokens=131,072. All other model scores are taken from the official leaderboard.<br>8. PaperBench: Evaluated in the BasicAgent setting under Code-Dev mode, judged by Claude Opus 4.6, and averaged over 3 runs (max 12 hours per run).<br>9. AndroidBench: Evaluated on the 95-task public subset, reporting avg@3 scores.<br>10. QwenSWEBench: Inhouse coding benchmark to evaluate 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-token context window.<br>11. QwenQoderBench: Inhouse coding benchmark to evaluate user experience on Qoder. Evaluated with the Claude Code harness. Reporting avg@5 with a 6-hour timeout, max_tokens=32,768, temperature=1.0, and a 256K-token context window.<br>12. QwenReactBench: Inhouse React project building benchmark using Claude Code as the harness, bilingual (EN/CN), 7 categories; auto-render + multimodal judge; BT/Elo rating.<br>13. QwenSVGBench: Inhouse SVG code generation benchmark; bilingual (EN/CN), auto-render + multimodal judge; BT/Elo rating.<br>14. CoWorkBench: Inhouse cowork benchmark for evaluating long-horizon tasks across computer science, finance, law, medical, and other productivity domains.<br>15. SkillsBench: Evaluated on the public SkillsBench v1.1 benchmark across 87 tasks, reporting the average score over three runs per task. Opus 4.8 and Fable 5 are evaluated on Claude Code; GPT-5.6 Sol is evaluated on Codex; the Qwen-series are evaluated on OpenCode. All results are from our own testing.<br>16. Automation-Bench: Evaluated on the 600-task public subset.<br>17. WideSearch: Evaluated with the Claude Code harness for external models and the Qwen-Agent harness for ours, reporting the average item-F1 over four runs.<br>18. $OneMillion-Bench: Evaluated using gemini-3.1-pro-preview.<br>19. PLawBench: Evaluated using gemini-3.1-pro-preview.<br>20. Empty cells (--): Scores are not yet available or are not applicable.</p>

</div>

API Usage

> [!Important]

> Qwen3.8-2.4T-A95B is a text-only model that requires thinking mode for all interactions. Multimodal inputs are not supported, and thinking cannot be disabled. Every response will automatically begin with reasoning enclosed in <think>\n...</think>\n\n before the final output.

> [!Tip]

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

> - temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, 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.

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.:

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

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/Qwen3.8-2.4T-A95B",
    messages=messages,
    extra_body={
        "chat_template_kwargs": {
            "enable_thinking": True,  # on by default; should not be turned off
            "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

    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

> [!Note]

> If you are using APIs from Qwen Cloud, in addition to changing model, please pass extra_body={"enable_thinking": True, "preserve_thinking": True} instead of extra_body={"chat_template_kwargs": {"enable_thinking": True, "preserve_thinking": True}}.

Best Practices

To achieve optimal performance, we recommend the following settings:

Sampling Parameters:

- We suggest using the following set of sampling parameters:

- temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, 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.

Citation

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

@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}
}

Run unsloth/Qwen3.8-2.4T-A95B-GGUF with guIDE

Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.

Download guIDE → · Browse 524k+ models · Compare models

Source: Hugging Face · Compare models