Qwen3.8-Max: Calling Alibaba's 2.4T Flagship Through an OpenAI-Compatible API

#ai #api #python #tutorial

Disclosure: I run NovAI, an API gateway that serves Qwen3.8-Max alongside 40+ other models. Specs below are from Alibaba's announcement; you can verify pricing yourself.

Alibaba's Qwen team released Qwen3.8-Max-Preview on 19 July 2026 at WAIC Shanghai: a 2.4-trillion-parameter sparse-MoE model, natively multimodal (text + images + video + documents), with a 128K context window (extendable to 1M). Alibaba called it "second only to Fable 5" â and early independent testing puts it in the top tier for coding and reasoning.

It's now available through an OpenAI-compatible endpoint. If you already use the OpenAI SDK, it's a two-line change.

TL;DR

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_KEY",
    base_url="https://aiapi-pro.com/v1",
)

resp = client.chat.completions.create(
    model="qwen3.8-max",
    messages=[{"role": "user", "content": "Explain sparse MoE routing in 3 bullet points."}],
)
print(resp.choices[0].message.content)

Streaming works exactly as you'd expect:

for chunk in client.chat.completions.create(
    model="qwen3.8-max",
    messages=[{"role": "user", "content": "Write a haiku about trillion-parameter models."}],
    stream=True,
):
    print(chunk.choices[0].delta.content or "", end="", flush=True)

Multimodal: images and documents

Qwen3.8-Max is natively multimodal â vision is built in, not bolted on. Send an image with the standard image_url content type:

resp = client.chat.completions.create(
    model="qwen3.8-max",
    messages=[{
        "role": "user",
        "content": [
            {"type": "text", "text": "What's in this screenshot? List all UI elements."},
            {"type": "image_url", "image_url": {"url": "https://example.com/ui.png"}},
        ],
    }],
)

It also accepts PDF, HTML, and Markdown directly â no pre-processing to plain text needed.

Why Qwen3.8-Max matters

Spec	Value
Total parameters	2.4T (sparse MoE)
Modality	Text, images, video, documents
Context window	128K standard (1M extended)
API protocol	OpenAI + Anthropic compatible
Open weights	Promised "soon" (no date yet)

An independent reviewer (Thomas Wiegold) ran it through 4 coding tests â including a full Go poker simulation that only Fable 5 and Grok 4.5 had previously one-shotted. Qwen3.8-Max became the third model to do it. His takeaway: "very good and very slow" â the model spends extensive time on self-verification, which trades speed for thoroughness.

Pricing

Alibaba's official rate for Qwen3.8-Max is Â¥12 / Â¥36 per million tokens (input / output) â roughly $1.67 / $5.00. During the preview period, Alibaba offers it at 1/10th of standard rate (Â¥1.2 / Â¥3.6, ~$0.17 / $0.50), with overnight rates as low as 1/50th (22:00â08:00 UTC+8).

NovAI passes these preview savings through. Check live per-token pricing: https://aiapi-pro.com/pricing

New keys come with $2 free credit, no credit card, so you can test Qwen3.8-Max against your own prompts before spending anything.