Skip to content

Qwen Coder models

Qwen's code-specialised models: tiny autocomplete models, a fast 30B MoE and Qwen3-Coder-Next, an 80B MoE that activates only 3B parameters per token.

  • Open weights
  • Models 3
  • Context up to 256K
  • Updated Nov 2026
Hugging Face Compare
On this page

At a glance

Type
Open-weight models
Developer
Alibaba Cloud (Qwen team) (China)
First release
Apr 2024
Models tracked
3 (1 current)
Longest context
256K tokens
Abilities
Tool calling
Licence
Apache 2.0, so commercial use is allowed.
Smallest local option
Qwen2.5-Coder 0.5B · ~1 GB
Website
qwen.ai
Hugging Face
huggingface.co/Qwen

Newest model

Qwen3-Coder-Next

Released Jan 2026 · 256K context

Smallest option

Qwen2.5-Coder 0.5B

~1 GB VRAM · 398 MB download

Qwen Coder models and versions

ModelReleasedSizesContextAbilitiesLicenceStatus
Qwen3-Coder-Next Agentic coding with 256K context on one large GPUJan 2026 80B 80B total, 3B active (MoE)256KToolsApache 2.0 Current
Qwen3-Coder Fast MoE coder that fits a 24 GB GPUJul 2025 30B · 480B 30B (3B active), 480B (35B active)256KToolsApache 2.0 Previous generation
Qwen2.5-Coder Small sizes are still the go-to for tab autocompleteSep 2024 0.5B · 1.5B · 3B · 7B · 14B · 32B 0.5B–32B32KToolsApache 2.0 Previous generation

Sizes, context windows and abilities come from the Ollama library and Hugging Face model cards, checked 14 Nov 2026. Release months are the first public upload. Spotted a mistake? Tell us.

How much VRAM does Qwen Coder need?

Download sizes per Ollama build and the GPU memory each needs with an 8K context. On a CPU-only server the same number is the RAM you need. Longer contexts need more memory; work it out with the LLM VRAM calculator.

Model4-bit (default)8-bit16-bitVRAM (4-bit)Fits onCommand
Qwen3-Coder-Next52 GB85 GB—~58 GB80 GB GPU
Qwen3-Coder 30B19 GB32 GB61 GB~21.5 GB24 GB GPU
Qwen3-Coder 480B290 GB510 GB960 GB~319.5 GBMulti-GPU
Qwen2.5-Coder 0.5B398 MB531 MB994 MB~1 GB8 GB GPU
Qwen2.5-Coder 1.5B986 MB1.6 GB3.1 GB~2 GB8 GB GPU
Qwen2.5-Coder 3B1.9 GB3.3 GB6.2 GB~3 GB8 GB GPU
Qwen2.5-Coder 7B4.7 GB8.1 GB15 GB~6 GB8 GB GPU
Qwen2.5-Coder 14B9 GB16 GB30 GB~10.5 GB12 GB GPU
Qwen2.5-Coder 32B20 GB35 GB66 GB~22.5 GB24 GB GPU

VRAM = 4-bit download + about 10% runtime overhead + 0.5 GB for an 8K context, rounded up. Mixture-of-experts (MoE) models still need all their weights in memory, but they run much faster than a dense model of the same size, so they also work well split between GPU and system RAM.

Run Qwen3-Coder-Next locally

# Install Ollama (Linux), then pull and chat
curl -fsSL https://ollama.com/install.sh | sh
ollama run qwen3-coder-next

New to Ollama? Follow Install Ollama on Ubuntu 26.04 (or 24.04) for GPU drivers, the systemd service and remote access.

What Qwen Coder is good for

IDE autocomplete

Qwen2.5-Coder 1.5B or 7B supports fill-in-the-middle and answers in milliseconds, even on a laptop.

Local coding agents

Qwen3-Coder 30B works with Continue, Cline, Aider and OpenCode on a single 24 GB card.

Repository-scale tasks

Qwen3-Coder-Next reads 256K tokens of code and was trained on 800K executable tasks.

Strengths and limits

Strengths

  • MoE designs (3B active) make generation fast even when the full model is large.
  • 256K context on Qwen3-Coder and Qwen3-Coder-Next.
  • Apache 2.0 on every size.

Watch out for

  • Qwen3-Coder-Next needs about 58 GB of memory at 4-bit.
  • Qwen3-Coder has no thinking mode; for hard debugging, a general model like Qwen3.8 can do better.

About Qwen Coder

Qwen Coder is the code-focused branch of the Qwen family. Qwen2.5-Coder (September 2024) set the standard for small open coding models; its 1.5B and 7B sizes are still the most common choice for editor autocomplete because they support fill-in-the-middle and run on almost anything.

Qwen3-Coder (July 2025) moved to mixture-of-experts: the 30B model activates only 3B parameters per token, so it generates quickly on a single 24 GB GPU, and the 480B model targets data centres. Qwen3-Coder-Next (January 2026) is built on Qwen3-Next-80B: it was trained with reinforcement learning on executable tasks and is aimed at coding agents that read whole repositories.

Guides for running Qwen Coder

Tool LLM VRAM Calculator: How Much GPU Memory Do You Need? Updated Nov 2026 AI & LLMs How to Install Ollama on Ubuntu 26.04 and Run LLMs Locally Updated Nov 2026 AI & LLMs How to Install llama.cpp on Ubuntu 26.04 (CPU and NVIDIA GPU) Updated Nov 2026 AI & LLMs How to Install vLLM on Ubuntu 26.04 (OpenAI-Compatible LLM Server) Updated Nov 2026 AI & LLMs How to Install Open WebUI on Ubuntu 26.04 (Self-Hosted ChatGPT Alternative) Updated Nov 2026

Frequently asked questions

What is the latest Qwen Coder model?

The newest Qwen Coder model is Qwen3-Coder-Next, released in Jan 2026. It has a 256K-token context window and supports tool calling.

How much VRAM do I need to run Qwen Coder locally?

The smallest option, Qwen2.5-Coder 0.5B, is a 398 MB download at 4-bit and needs about 1 GB of GPU memory (or RAM on a CPU-only machine) with an 8K context. Larger sizes need more; see the hardware table on this page, and use the VRAM calculator for longer contexts.

How do I run Qwen Coder with Ollama?

Install Ollama, then run "ollama run qwen2.5-coder:0.5b". Ollama downloads the 4-bit build and starts a chat; the same model is also available on the local API at http://localhost:11434.

Can I use Qwen Coder commercially?

Apache 2.0, so commercial use is allowed. Always read the licence of the exact model you deploy; it is linked in the models table.

Compare Qwen Coder

← All AI models

Back to top