Qwen3-Coder-Next · 80B A3B

Qwen3-Coder-Next is an open-weight causal language model from the Qwen/Qwen3-Coder-Next repository, licensed under apache-2.0 and tagged for text-generation with safetensors. It is designed specifically for coding agents and local development. The model has 80B total parameters with 3B activated parameters (79B non-embedding), a hidden dimension of 2048, 48 layers in a hybrid Gated DeltaNet/ Gated Attention + MoE layout, and a native context length of 262,144 tokens. It supports only non-thinking mode and does not generate thinking blocks. According to the model card, it is intended for text generation and offers advanced agentic capabilities for long-horizon reasoning and tool use. The card states compatibility with various CLI/IDE platforms and local applications, but provides no specific hardware requirements or performance numbers beyond parameter counts and context length.

Verified sourceRepository opened Jan 30, 2026Source checked 9/23/2026Version: a7fbcb5c

LOCALRENTED GPUOPEN WEIGHTS

At a glance

Parameters: 80B total / 3B active
Architecture: Hybrid attention MoE
Context length: 262,144 native
License: Apache 2.0
Disk space: 52 GB
Software: Ollama, vLLM, SGLang

View the model source

MY HARDWARE

Will it run on your machine?

Save your machine to see a personalized rating and its reasoning.

Add my hardware

Best for

Coding agents and local development tasks
Long-horizon reasoning with tool usage
Integration with CLI/IDE platforms via scaffold templates
Code generation from prompts
Use with local applications such as Ollama, LMStudio, MLX-LM, llama.cpp, and KTransformers

Tradeoffs

The model card states it supports only non-thinking mode and does not generate thinking blocks
Hardware requirements and inference performance are not specified in the provided facts
Must use factual model parameters: 80B total parameters and 3B activated parameters
Native context length is 262,144, but reducing context length may be needed to avoid out-of-memory issues according to the card
Only text-generation pipeline and safetensors tags are provided

Ways to run it

vLLM · Linux

Repository-specific example documented in the publisher model card; confirm dependencies and platform in the source.

vllm serve Qwen/Qwen3-Coder-Next --port 8000 --tensor-parallel-size 2 --enable-auto-tool-choice --tool-call-parser qwen3_coder

Official instructions

Transformers · See official guide

Repository-specific example documented in the publisher model card; confirm dependencies and platform in the source.

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "Qwen/Qwen3-Coder-Next"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
  model_name,
  torch_dtype="auto",
  device_map="auto"
)

# prepare the model input
prompt = "Write a quick sort algorithm."
messages = [
  {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
  messages,
  tokenize=False,
  add_generation_prompt=True,
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# conduct text completion
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=65536
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist() 

content = tokenizer.decode(output_ids, skip_special_tokens=True)

print("content:", content)

Official instructions