NEWS & VIEWS · AI MODELS

Alibaba launches Qwen4-Max, a cheaper flagship AI model aimed at coding and office work

Alibaba's newest flagship model costs a quarter less per token than the last one and lands in the top 10 on a popular coding leaderboard. Most of the scores are early, and the model isn't open source.

Qwen post on X announcing Qwen4-Max, with a bar-chart benchmark grid comparing it to other models.
Screenshot: Qwen (@Alibaba_Qwen) on X

TL;DR

Is Qwen4-Max worth trying?

Alibaba released Qwen4-Max, its strongest AI model so far, at a lower price than the last version. If you pay for AI by usage, strong coding results just got cheaper, but most scores are early or Alibaba's own. Try it on a small task before you move real work.

What is Qwen4-Max?

Alibaba announced Qwen4-Max on Sunday, November 15, 2026, at 9:01pm ET. It's the first model in the new Qwen4 series and the most capable Qwen model yet.

It can read images as well as text, and it has a 1-million-token context window. Tokens are the small chunks of text AI models read and bill by, and the context window is how much the model can hold in mind at once, so 1 million tokens covers a very large pile of documents or code. Alibaba aims it at office work, coding, design and long-running agents, meaning AI that works through multi-step tasks on its own.

Under the hood, Alibaba says it has 2.4 trillion total parameters (the internal settings a model learns) but uses only 71 billion active parameters for each token. That means only a slice of the model works on each step, which helps keep costs down.

Alibaba also says Qwen4-Max beats Qwen3.8-Max while using fewer tokens, cutting average cost per task by more than 60% by its own measure.

How much does Qwen4-Max cost?

These are Alibaba's own API prices. An API is the connection developers use to plug a model into their apps, paid by usage.

Qwen4-Max prices from Alibaba's launch post and blog. Qwen3.8-Max prices from Artificial Analysis. Resellers may charge differently.
Per million tokensQwen4-MaxQwen3.8-Max
Input (what you send)US$1.50US$2.00
Output (what it writes)US$4.50US$6.00
Cached input (repeat text it already saw)US$0.075US$0.25

That's 25% less for input and output, and a much bigger cut on cached input, which matters if your app sends the same long instructions again and again.

How good is it?

Early numbers look strong, but treat them as early:

  • Code Arena: WebDev: 6th place with 1,750 points. Arena marks the score as preliminary.
  • Artificial Analysis Intelligence Index: 43, which is 9 points above Qwen3.8-Max and 3 points behind Kimi K3.1 Pro. It costs about US$3.27 per index task, roughly 11% less than Kimi K3.1 Pro.

The 60% cost-per-task cut is Alibaba's own figure. Bloomberg's headline framed the launch as Alibaba's "riposte to US rivals."

Where can you use it?

Qwen4-Max is live now in Alibaba's QwenCloud API and Token Plan, plus Qwen Studio, QwenWork and Qoder. It also went live the same night on OpenRouter, Vercel AI Gateway and OpenCode, so many developers can switch to it without a new account.

It isn't open source. Alibaba hasn't released the model's weights, the trained files you'd need to run it yourself. It says open weights for smaller Qwen4 models are coming, with no date.

X reactions

Benchmark trackers and model hubs moved fast on Sunday night:

Our take

For anyone paying for AI by the token, such as small app builders, agencies and automation shops, this is good news. A model that ranks near the top on coding, at US$1.50 in and US$4.50 out per million tokens, puts more pressure on everyone's prices. It also shows the gap between US and Chinese labs is narrow.

We'd still go slowly. The scores are early, the biggest savings claim comes from Alibaba's own numbers, and resellers may price it differently. Before you send client data anywhere, check where that provider stores it and whether your company or your clients allow it. Your wallet will like the price, but your contracts get the final say.

— Artificially Nerdy

What to do this week

  1. Run one real task side by side with the model you use now, and compare quality and cost.
  2. Check your provider's price page. OpenRouter and Vercel AI Gateway may not match Alibaba's own rates.
  3. Ask where data goes before sending client or customer files, and check your company's rules on AI providers.
  4. Use a separate API key for trying new models, and lock down your AI API keys so a leak can't run up a bill.
  5. Waiting for open weights? Watch for the smaller Qwen4 models Alibaba has promised.

FAQ

Can I use Qwen4-Max for free?

Alibaba lists paid API pricing and its Token Plan. It hasn't announced a free tier for Qwen4-Max.

Can I run Qwen4-Max on my own computer?

No. The weights aren't public. Only smaller Qwen4 models are promised as open weights.

Is Qwen4-Max better than Qwen3.8-Max?

On the Artificial Analysis Intelligence Index it scores 9 points higher, and Alibaba says it beats Qwen3.8-Max while using fewer tokens.

Can Qwen4-Max read images?

Yes. It has native image understanding as well as text.

Sources

  1. Alibaba Cloud blog: Alibaba unveils Qwen4-Max — November 16, 2026 (Hangzhou). Parameters, uses, cost-per-task claim, cache price, Code Arena rank, open-weights plan.
  2. Qwen blog: Qwen4-Max
  3. QwenCloud: Qwen4-Max model page
  4. Alibaba Qwen launch post on X — November 15, 2026, 9:01pm ET. API prices.
  5. Bloomberg: Alibaba Unveils New Flagship AI Model in Riposte to US Rivals — November 16, 2026. Headline only.
  6. Artificial Analysis on X — November 15, 2026, 9:55pm ET. Index score, cost per task, Qwen3.8-Max prices.
  7. OpenRouter on X — November 15, 2026, 9:41pm ET.

THE NERD SIGNAL

Rocky digs through the AI noise. You get the good bits.