Qwen 3.8

Generic Info

• Publisher: Alibaba Cloud
• Release Date: 2026
• Model Family: Dense (Qwen3.8-27B) through flagship MoE (Qwen3.8-Max)
• Context Window: 128K standard, extendable up to 1,000,000 tokens (1M context)
• License: Apache 2.0
• Key Capabilities: Native Image/Video Understanding, Flexible Thinking Control (`xhigh`, `medium`, `low`), 119 Languages, Built-in Tool Calling

Building on the massive global adoption of the Qwen series, Qwen 3.8 is the most capable generation in the family to date. Engineered as a native vision-language model, Qwen 3.8 understands complex diagrams, multi-frame video, and dense multilingual text. Its revolutionary thinking control allows developers to dial the reasoning effort from low latency up to "xhigh" for deep mathematical proofs, competitive programming, and long-horizon tool execution.

Hello World Guide

Run Qwen 3.8 locally with Hugging Face transformers or vLLM.

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "Qwen/Qwen3.8-27B"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto"
)

messages = [
    {"role": "system", "content": "You are a helpful assistant with deep reasoning and tool use capabilities."},
    {"role": "user", "content": "Prove that the square root of 2 is irrational, providing step-by-step mathematical rigor."}
]

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=True,      # Toggle thinking mode
    reasoning_effort="xhigh"   # Options: 'xhigh', 'medium', 'low'
)

model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=1024
)

response = tokenizer.decode(generated_ids[0][model_inputs.input_ids.shape[1]:], skip_special_tokens=True)
print(response)

Industry Usage