Alibaba releases open weights for Qwen3. 8 AI model family

Qwen3. 8 Alibaba open weights multimodal AI large language models

Alibaba has released the open weights for its new Qwen3. 8 AI model family under the Apache 2.0 license. The release includes the Qwen3. 8-27B, a 27-billion-parameter multimodal model designed for local deployment on consumer-grade hardware, such as GPUs with 24GB of VRAM. This model supports text, image, and video inputs and features a native context window of 262K tokens, which can be extended to 1 million tokens using YaRN technology.

Additionally, Alibaba introduced the Qwen3. 8-Max (also referred to as Qwen3. 8-2.4T-A95B), a flagship model utilizing a Mixture of Experts (MoE) architecture. While it contains 2.4 trillion total parameters, it only activates 95 billion parameters per request to maintain efficiency. This model is intended for enterprise-scale workloads and requires significant multi-GPU infrastructure.

Alibaba reports that its Qwen ecosystem has achieved significant scale, with over 3 billion downloads globally in the last six months, surpassing competitors like Meta and Google on the Hugging Face platform. Benchmarks provided by Alibaba suggest the 27B model outperforms previous versions in coding and office tasks and competes with Anthropic’s Opus 4. 6 in several categories, though it trails in pure reasoning tasks.

Entities

Alibaba · Alibaba Group Holding · Anthropic · Hugging Face · Meta Platforms · Motif Technologies · Nvidia · OpenAI · Qwen · White House

Claims

What the coverage asserts, and how many sources carry each claim.

[● 7 SOURCES] Alibaba released open weights for the Qwen3. 8-27B multimodal model. cryptobriefing.com · www.ithome.com · techgenyz.com · thenewstack.io · platum.kr · +2 more
[● 3 SOURCES] The Qwen3. 8-Max model features 2.4 trillion total parameters with 95 billion active parameters using a Mixture of Experts architecture. cryptobriefing.com · techgenyz.com · ebizlatam.com
[○ 1 SOURCE] Alibaba claims the Qwen3. 8-27B model performs at a level comparable to Anthropic’s Opus 4. 6. thenewstack.io
[● 3 SOURCES] The Qwen3. 8-27B model supports a 262K native context window, extendable to 1M tokens via YaRN technology. www.ithome.com · 4sysops.com · www.blocktempo.com
[● 4 SOURCES] The Qwen3. 8 series shows performance improvements over the previous Qwen 3. 7-Plus in coding and knowledge work. www.ithome.com · thenewstack.io · 4sysops.com · cryptobriefing.com
[● 2 SOURCES] The Qwen3. 8-27B model is multimodal, capable of processing text, image, and video inputs. cryptobriefing.com · 4sysops.com
[● 2 SOURCES] The Qwen3. 8-Max model features 2.4 trillion parameters and a context window of up to 1 million tokens. ebizlatam.com · techgenyz.com
[○ 1 SOURCE] In official comparisons, the Qwen3. 8-27B model outperformed Claude Opus 4. 6 Max in 9 categories but trailed in 5 categories related to pure reasoning. www.blocktempo.com
[○ 1 SOURCE] Alibaba's open-source AI models reached over 3 billion downloads globally in the last six months. actualidad.rt.com
[○ 1 SOURCE] The Qwen ecosystem includes over 460 models and has generated more than 300,000 community-created derivatives. actualidad.rt.com
[● 2 SOURCES] Alibaba released the Qwen3. 8-27B model, a 27 billion parameter multimodal model under Apache 2.0 license. www.blocktempo.com · dev.to