Qwen 3.8 27B

Model ID: vllm/release/qwen3.8-27b Qwen (Alibaba) Productive

EU AI Act

Attribute	Value
Classification	GPAI (General-Purpose AI Model)
Risk class	Limited Risk (Article 50)
Transparency obligation	Yes
noris role	Distributor

As a GPAI model, this model falls within the scope of the EU AI Act transparency obligations. noris acts as a distributor and makes the model developer’s information available pursuant to Article 53(2) and (3).

Technical Specifications

Property	Value
Model name	Qwen 3.8 27B
Provider	Qwen (Alibaba)
Model ID	vllm/release/qwen3.8-27b
Parameters	27B (dense)
Architecture	Hybrid (Gated DeltaNet + Gated Attention), MTP
Context length	262,144 native, up to 1,000,000 via YaRN
Modalities	Text + image + video (multimodal)
Checkpoint	NVFP4, quantized by Inferact (Model Optimizer)
License	Apache 2.0

Recommended Use Cases

• Agentic coding (Terminal Bench 2.1: 73.0%, SWE-bench Pro: 61.7%)
• Computer and browser use (OSWorld-Verified: 84.3%)
• Image and video understanding, incl. hour-scale videos
• Long-horizon tasks and office workflows (CoWorkBench: 70.7%)

Known Limitations

Caution

• Thinking mode enabled by default (generates extra tokens): enabled thinking mode increases token consumption and latency. Disable via chat_template_kwargs.enable_thinking=false, tune depth with reasoning_effort (xhigh, medium, low).
• YaRN beyond 262K context: contexts exceeding 262,144 tokens require RoPE scaling (YaRN) and may degrade performance on short texts.
• Primarily EN/ZH: performance is mainly optimized for English and Chinese.

Risk Assessment (EU AI Act)

Risk Factor	Category	Level
Open source (Apache 2.0)	Misuse potential	Low
Multimodal	Image and video input, deepfake potential	Medium
Environmental impact	Model size (27B dense)	Medium

Training Data & Sustainability

Training data: not fully publicly documented.

Sustainability:

Metric	Value
CO₂	Medium (27B dense)
Optimization	Hybrid architecture (Gated DeltaNet) for efficient long contexts

Original Sources

• Base model: https://huggingface.co/Qwen/Qwen3.8-27B
• NVFP4 checkpoint: https://huggingface.co/Inferact/Qwen3.8-27B-NVFP4
• License: Apache 2.0

Model card created on October 7, 2026. This model card serves EU AI Act transparency purposes (Regulation (EU) 2024/1689). The information is based on publicly available data from the model provider.