Qwen 3.8 27B
Apache 2.0 dense multimodal model designed for local deployment on consumer hardware
Context window
262K
Input / 1M tokens
Free
Output / 1M tokens
Free
Provider
Alibaba
Data verified 2026-08-18
Qwen 3.8 27B is a 27-billion parameter dense multimodal model released under Apache 2.0 license. It features native image and video understanding, a 262,144-token context window (extendable to 1M via YaRN), and flexible reasoning control. Positioned for local deployment and edge AI on consumer GPUs and workstations.
Capability index
Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.
How to access it
Open-weight model available on Hugging Face and ModelScope. Download weights directly from Qwen/Qwen3.8-27B repository or use quantized variants through community packages.
Strengths
- ✓Runs on consumer hardware (16-17GB VRAM at quantized precision)
- ✓Native multimodal: text, image, and video understanding
- ✓Strong performance on coding and office automation tasks
- ✓Open-source with permissive Apache 2.0 license
- ✓Substantial improvements over Qwen 3.6-27B
Best for developers who...
When to choose it (and when not to)
Reach for Qwen 3.8 27B when...
- →When you need to run AI locally without cloud dependency
- →For coding and agent use cases
- →When data privacy is critical
- →For teams with limited GPU infrastructure
Look elsewhere if...
- ✕If you need the largest reasoning capabilities (use Qwen 3.8-Max instead)
- ✕For tasks requiring extreme throughput at scale
How to use it
- ›Configure reasoning_effort dial to control reasoning depth
- ›Works well for tool-calling and JSON output
- ›Supports OpenAI-compatible Chat Completions API
Quickstart
Pythonfrom transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained('Qwen/Qwen3.8-27B')
model = AutoModelForCausalLM.from_pretrained('Qwen/Qwen3.8-27B', device_map='auto')For quantized inference, use community GGUF builds or LM Studio. OpenRouter offers hosted endpoint at $0.45/$3.20 per 1M tokens.
API model id: Qwen/Qwen3.8-27B
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| SWE-Bench Pro | 61.7% | Software engineering benchmark - Alibaba evaluation |
| DeepSWE 1.1 | 42.2% | Significant improvement from predecessor's 13.3% |
| QwenSWEBench | 79.0% | |
| LiveCodeBench v6 | 90.3% | |
| GPQA Diamond | 89.2% | |
| CoWorkBench | 70.7% | Office productivity tasks |
Source: Official Qwen 3.8 27B Model Card
Compare Qwen 3.8 27B
Compare Qwen 3.8 27B with any other model
Build a comparison →All model comparisons →