Qwen3.8 27B

What this model is

The architecture is dense. Every parameter is part of the active path, so memory and speed scale with the full model rather than a smaller expert subset.

Memory and quantization

Decimal GB. Sizes combine the published GGUF weight file with the F16 multimodal projector required for image and video input.

Estimate formula: published weight files + 0.5369 GB per 8K tokens × context / 8K + 1.5 GB runtime overhead. Sizes are decimal GB, not GiB. At 8K context, the smallest listed build is about 19.4 GB before OS and application reserve.

Full-attention KV estimate from the public config. Sixteen of 64 layers use full attention; linear-attention state is small and excluded from this conservative per-token figure.

KV method source: https://huggingface.co/Qwen/Qwen3.8-27B/raw/main/config.json

Speed heuristic

Not benchmarked here. Dense 27B inference is limited by all 27B parameters on every token.

Active path: 27B parameters. This is a relative speed signal only; it is not a tokens/sec benchmark.

Runtimes

• Ollama
• llama.cpp
• MLX
• vLLM

ollama run qwen3.8:27b

License restrictions

Apache 2.0 permits commercial use and modification; retain notices and comply with the patent and termination terms.

This is a plain-language summary, not legal advice. Read the linked license before commercial deployment or redistribution.

Known issues

• A Q4 build plus KV cache is close to a 20 GB memory floor, so 16 GB devices need a smaller model or aggressive offload.
• The GGUF package needs the multimodal projector for vision and video input.

Sources

• https://huggingface.co/Qwen/Qwen3.8-27B
• https://huggingface.co/unsloth/Qwen3.8-27B-GGUF
• https://ollama.com/library/qwen3.8
• https://huggingface.co/Qwen/Qwen3.8-27B/raw/main/config.json

This page is an independent compatibility guide, not official documentation. Verify the primary source before downloading or deploying a model.

Check a specific machine