Qwen3.8-Max
Qwen3.8-Max2.4T MoE reasoning model, first open-weight Max-scale model, multimodal (text/image/video), 1M context, top-10 AA Intelligence.
Model details
Qwen3.8-Max is Alibaba's latest frontier reasoning model, released August 3, 2026, and the most capable entry yet in the Qwen Max series. It's a 2.4T parameter MoE model with 95B active parameters, and it's the first open-weight model at Max scale, with weights coming to Hugging Face and ModelScope. It scores 58 on the Artificial Analysis Intelligence Index — a 23% jump from its predecessor, Qwen3.7-Max — placing it in the top 10 of all benchmarked models. It's also the first Qwen Max model to go multimodal, accepting text, image, and video input alongside a 1M token context window.
The model combines configurable reasoning depth, function calling, and JSON mode, making it well-suited for agentic workflows, complex document analysis, visual reasoning tasks, and coding. It's notably verbose in its reasoning output, trading higher token usage for stronger answer quality.
License note: Qwen3.8-Max uses a custom license. Commercial use is permitted, but high-revenue Model-as-a-Service providers (>$50M annually) require a separate license from Qwen. Review the license before deploying.
1import os
2from openai import OpenAI
3
4# For a Dedicated Inference deployment, use your model's dedicated endpoint
5model_id = "YOUR_MODEL_ID" # from your Baseten dashboard
6model_url = f"https://model-{model_id}.api.baseten.co/environments/production/sync/v1"
7
8client = OpenAI(
9 base_url=model_url,
10 api_key=os.environ["BASETEN_API_KEY"],
11)
12
13response = client.chat.completions.create(
14 model="Qwen/Qwen3.8-Max", # must match --served-model-name in the deployment
15 messages=[
16 {"role": "system", "content": "You are a helpful assistant."},
17 {"role": "user", "content": "Summarize the key differences between Qwen3.7 Max and Qwen3.8 Max."},
18 ],
19 max_tokens=2048,
20 stream=True,
21 # Qwen3.8 Max-specific reasoning controls, passed via extra_body
22 extra_body={
23 "reasoning_effort": "xhigh", # "xhigh" (default) | "medium" | "low"
24 "preserve_thinking": True, # keep reasoning tokens in the response
25 },
26)
27
28for chunk in response:
29 if chunk.choices[0].delta.content is not None:
30 print(chunk.choices[0].delta.content, end="")
31
