Skip to content

Qwen3.6 vs Qwen3.8

Qwen3.6 and Qwen3.8 side by side: model sizes, the VRAM you need, context window, abilities, licence and which one to pick.

Side by side

Qwen3.6Qwen3.8
TypeModel in the Qwen familyModel in the Qwen family
DeveloperAlibaba Cloud (Qwen team)Alibaba Cloud (Qwen team)
ReleasedApr 2026Aug 2026
Parameters27B dense, 35B MoE (3B active)27B
Sizes27B · 35B27B
Smallest VRAM (4-bit)~20.5 GB~20.5 GB
Context window256K tokens256K tokens
AbilitiesTool callingReasoning modeImage inputTool callingReasoning modeImage input
LicenceApache 2.0Apache 2.0
Run itollama run qwen3.6:27bollama run qwen3.8:27b

Which one should you pick?

Pick Qwen3.6 if…

  • your tools or fine-tunes already target it (released Apr 2026, so support is mature)

Best for: Agentic coding; the 35B MoE is very fast

Pick Qwen3.8 if…

  • you want the newer model (Aug 2026)

Best for: Flagship local model for coding, agents and vision

VRAM and download size

Ollama's default 4-bit builds with an 8K context. Read GB as system RAM on a CPU-only machine or a Mac.

Qwen3.6

ModelDownloadVRAM
27B18 GB~20.5 GB
35B23 GB~26 GB

Qwen3.8

ModelDownloadVRAM
27B18 GB~20.5 GB

Frequently asked questions

Is Qwen3.6 better than Qwen3.8?

It depends on the job and your hardware. For quality, test both on your own prompts: benchmark rankings change with every release.

Which needs less VRAM, Qwen3.6 or Qwen3.8?

Qwen3.6 starts at ~20.5 GB (18 GB download). Qwen3.8 starts at ~20.5 GB (18 GB download).

Can I use Qwen3.6 and Qwen3.8 commercially?

Qwen3.6: Apache 2.0. Qwen3.8: Apache 2.0. Check the exact licence of the model you deploy; custom licences can limit very large companies or certain uses.

More comparisons

← All AI models

Back to top