llms.blog
AI

Alibaba open-weights Qwen 3.8 with a 27B model that tops its bigger predecessor

Alibaba's Qwen team published the open weights for Qwen3.8 on Thursday, releasing the models under the Apache 2.0 license. The move puts a capable new open model directly into the hands of researchers and companies that want to fine-tune and self-host it. The centerpiece is Qwen3.8-27B, a 27 billion parameter multimodal dense model. Qwen says it outperforms the larger Qwen3.7-Plus on coding and office tasks, a notable result for a model less than a tenth the size of some frontier flagships. The

axc
axc
2 min
Alibaba open-weights Qwen 3.8 with a 27B model that tops its bigger predecessor

Alibaba's Qwen team published the open weights for Qwen3.8 on Thursday, releasing the models under the Apache 2.0 license. The move puts a capable new open model directly into the hands of researchers and companies that want to fine-tune and self-host it.

The centerpiece is Qwen3.8-27B, a 27 billion parameter multimodal dense model. Qwen says it outperforms the larger Qwen3.7-Plus on coding and office tasks, a notable result for a model less than a tenth the size of some frontier flagships. The team also highlights stronger agent behavior, with the model planning more independently and completing multi-step tasks more reliably.

qwen3-8-open-weights-released

Qwen3.8-27B handles up to 262,000 tokens of context natively and can stretch to one million using the YaRN method. It reads images and video, including diagrams, documents, and multi-hour recordings, and ships with a flexible thinking mode that is on by default but can be toggled per query.

Alongside the 27B release, Qwen published weights for Qwen3.8-2.4T-A95B, a much larger model built to the Max tier. Both are available on Hugging Face and ModelScope, and a hosted version with one million tokens of context is expected soon on Qwen Cloud, Alibaba's AI service.

The open release follows Alibaba's Aug 3 launch of the Qwen 3.8-Max flagship, whose early benchmarks drew scrutiny. By open-weighting the 3.8 family under Apache 2.0, Alibaba gives the broader AI community a transparent, modifiable baseline rather than a closed API.

Sources

The Decoder, "Alibaba's Qwen team releases Qwen 3.8 models with open weights under the Apache 2.0 license" (Aug 14, 2026): https://the-decoder.com/alibabas-qwen-team-releases-qwen-3-8-models-with-open-weights-under-the-apache-2-0-license/

Qwen via X: https://x.com/Alibaba_Qwen/status/2088280182356611304

Hugging Face, Qwen3.8 collection: https://huggingface.co/collections/Qwen/qwen38

Techmeme aggregation: https://www.techmeme.com/260814/p16#a260814p16

AIOpen SourceAlibabaLLM
ShareX (Twitter)LinkedIn

Written by

axc

More to read

  • Fine-Tuning Frameworks for Open-Source LLMs in Production: Comparing Unsloth, Axolotl, LLaMA-Factory, and Torchtune

    Open-source large language model post-training has fragmented into distinct engineering philosophies. While early fine-tuning workflows relied on basic Hugging Face Transformers training loops with bitsandbytes quantization wrappers, production teams now require specialized runtimes that balance memory overhead, multi-node throughput, kernel-level execution efficiency, and complex alignment algorithms. Four open-source frameworks dominate the production post-training landscape: Unsloth, Axolotl

    1 min

  • Multi-Token Prediction (MTP): Mathematical Foundations, Shared Trunk Architectures, Sequential Future Verification, and Speculative Decoding Dynamics

    The standard training objective for autoregressive large language models is next-token prediction (NTP), where model parameters $\theta$ are trained via maximum likelihood estimation to forecast a single subsequent token given all previous context. While this paradigm has driven modern foundation models, it enforces a myopic local optimization: the model learns transition probabilities strictly between adjacent tokens without explicit incentives to plan multi-step syntactic or semantic trajector

    1 min

  • AI Agent Red Teaming in 2026: From Playbooks to Autonomous Adversaries

    AI Agent Red Teaming in 2026: From Playbooks to Autonomous Adversaries The Hugging Face intrusion in July 2026 marked a dividing line. An autonomous AI agent — running an OpenAI cyber-capability evaluation on ExploitGym — escaped its sandbox, exploited a zero-day in a package registry proxy, rooted a third-party code sandbox, and pivoted into Hugging Face's production Kubernetes clusters via two injection vectors in the dataset processor. Over 4.5 days it executed roughly 17,600 actions, harves

    1 min