Qwen 3.8 27B

Overview

Alibaba's Qwen 3.8 27B is a dense, multimodal model with native vision support and a 262K context window, released under Apache 2.0 with notable gains in agentic coding.

When to pick this model

• Agentic coding tasks needing strong tool-use and long-horizon planning
• Multimodal workflows involving images, documents, or video
• Long-context document or codebase analysis (up to 262K natively, 1M via YaRN with vLLM/SGLang)
• Local deployment on Apple Silicon via MLX or via Ollama

VRAM requirements by quantization

Quantization	VRAM required
Q4_K_M (recommended)	16 GB
Q5_K_M	19 GB
Q8_0	29 GB
FP16 (no quantization)	54 GB

VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.

In practice, Qwen 3.8 27B needs a 16 GB card at Q4_K_M (16 GB). Stepping up to Q8_0 nearly doubles the footprint to 29 GB, and unquantized FP16 weights take 54 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.

Without a GPU, Qwen 3.8 27B needs roughly 35 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 9 tokens/sec on entry-level GPUs, on the order of 14 tokens/sec on a mid-range card, and up to 22 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.

What hardware do you need

The table below matches Qwen 3.8 27B to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.

GPU memory	Example cards	Best fit for Qwen 3.8 27B
8 GB	RTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GB	Does not fit — needs 16 GB at Q4_K_M
12 GB	RTX 5070, RTX 5070 Ti Laptop, RTX 4080 Laptop	Does not fit — needs 16 GB at Q4_K_M
16 GB	RTX 5080, RTX 4080 Super, Radeon RX 9070 XT	Q4_K_M (16 GB used)
24 GB	RTX 4090, Radeon RX 7900 XTX, RTX 5090 Laptop	Q5_K_M (19 GB used)
32 GB	RTX 5090	Q8_0 (29 GB used)

Which hardware should you buy to run Qwen 3.8 27B?

To run Qwen 3.8 27B locally at Q4, you need ~16 GB of VRAM. The best value for this today is a RTX 5070 Ti 16GB (GIGABYTE Gaming OC) (16 GB VRAM).

Published benchmark scores

Benchmark	Score
Terminal Bench 2.1	73
SWE-bench Pro	61.7
LiveCodeBench v6	90.3
GPQA Diamond	89.2
OSWorld-Verified	84.3

Scores published by the model author or aggregated from public leaderboards. Re-measured monthly by our editorial team.

Strengths

• Unrestricted Apache 2.0 license
• Native vision support for images, documents, and video
• 262K native context
• Clear agentic coding improvement over Qwen 3.6-27B
• Official Ollama tag plus MLX builds for Apple Silicon

Limitations

• 18GB of Q4_K_M weights means a 24GB GPU is needed to stay fully on-GPU
• Modest local throughput (~9-22 tok/s), and thinking mode adds token overhead
• Benchmarks are vendor-reported, not independently reproduced
• The 1M-token context mode requires vLLM/SGLang and isn't available through Ollama

Typical workloads

In our catalog grid, Qwen 3.8 27B is filed under Agentic Code, Local Multimodal, Long Context — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: code generation and review (pair it with an editor integration like Continue.dev or Cline); multi-step reasoning and math-flavoured tasks; vision-language work — screenshots, charts, scanned documents; multilingual workloads.

The 256k-token context window is large enough to hold entire codebases' worth of files or long reports in a single prompt, which is what makes local RAG and document analysis practical. The Apache 2.0 license is permissive, so shipping it inside a commercial product raises no special legal questions.

Architecture & training

Architecture: Dense 27B · hybrid attention (64 layers: Gated DeltaNet + Gated Attention) · native vision-language (image and video) · 262,144 token context, extendable to 1M via YaRN

Training: Qwen 3.8 generation (Alibaba), a dense open-weight variant released on August 14, 2026. Trained with Multi-Token Prediction. "Thinking" mode is active by default, with depth adjustable via reasoning_effort.

Quick start

ollama run qwen3.8:27b

Or use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.

Similar models worth comparing

Frequently asked questions

How much VRAM does Qwen 3.8 27B need?

At the recommended Q4_K_M quantization, Qwen 3.8 27B needs about 16 GB of VRAM. Q8_0 takes 29 GB, and unquantized FP16 weights take 54 GB.

Can Qwen 3.8 27B run without a GPU?

Yes — with roughly 35 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.

What context window does Qwen 3.8 27B support?

Qwen 3.8 27B supports a 256k-token context window (262,144 tokens).

Can I use Qwen 3.8 27B commercially?

Yes. Qwen 3.8 27B is released under Apache 2.0, a permissive open-source license that allows commercial use, modification and redistribution.

How fast is Qwen 3.8 27B on consumer hardware?

Our compatibility engine estimates on the order of 14 tokens/sec on a mid-range GPU and up to 22 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.

Which quantization of Qwen 3.8 27B should I download first?

Start with Q4_K_M (16 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. On a 24 GB card you can run up to Q5_K_M.