Alibaba Launches Qwen4-Max, a Flagship AI Model With Lower API Prices

Alibaba Launches Qwen4-Max, a Flagship AI Model With Lower API Prices

Alibaba has released Qwen4-Max, a new flagship AI model that the company pitches as a round-the-clock teammate for coding, office work and long-running agent tasks. The launch on November 16, 2026 also brings a price cut, with list rates in Singapore sitting 25% below the previous Qwen3.8-Max. It is the first model in the new Qwen4 family, and it lands in the middle of a busy stretch of releases from Chinese AI labs.

What Qwen4-Max Is and What Changed

Qwen4-Max is built on a new architecture that Alibaba designed for the whole Qwen4 series. The company says the model has a 2.4-trillion-parameter backbone, but only about 71 billion of those parameters switch on for each token it processes. That design, known as a mixture of experts, is meant to keep a very large model affordable to run.

The model reads text, images and video and replies in text. It also keeps a one-million-token context window, which is the amount of material it can work with in a single request. Both of those features were already offered by the previous flagship, so the main changes are under the hood.

Built for Long Tasks

Alibaba is leaning hard on the idea of agents that work for hours without supervision. A new feature called Live Delivery has the model hand over a usable first version early, then improve it in stages while you steer. The company is also open-sourcing a plugin so developers can try that workflow themselves.

The launch post shows the model building 3D scenes in Blender, making playable games in Unity and Unreal Engine, and producing financial research with traceable sources. Those are the company’s own demos, so treat them as a preview and not as independent proof.

Qwen4-Max Pricing and Benchmarks

The headline for developers is cost. Alibaba says Qwen4-Max does better work than its predecessor while using fewer tokens, and the per-token price is lower too. Here is how the main numbers compare, based on the company’s own figures and its pricing pages.

ItemQwen3.8-MaxQwen4-Max
Work360 score (Alibaba’s own benchmark)47.7452.69
Average cost per Work360 task$4.43$1.67
Input price per million tokens (Singapore)25% higher$1.50
Output price per million tokens (Singapore)25% higher$4.50
Secure code pass rate (function level)22.0%31.8%

Work360 is a test Alibaba created to measure professional work across 26 broad industries, so the score is not a neutral yardstick. The company also reports roughly 48% to 49% fewer tokens on the Terminal-Bench 4.0 coding test. On Work360, token use fell by about 24% to 35%, depending on the reasoning setting.

Alibaba also claims the model scored near the top of coding benchmarks, ahead of Anthropic’s Claude Fable 5.1 and Moonshot’s Kimi K3.1 Pro, according to Bloomberg. Those are company claims, and independent testing will take time to confirm or challenge them. If you already rely on Claude models for coding, a real comparison on your own tasks matters more than any chart.

Not Open Weight, at Least for Now

The Qwen family is best known for models you can download and run yourself. Qwen4-Max breaks from that, because it is available only as a hosted service. Alibaba says smaller Qwen4 models will be released as open source, but it gave no date, sizes or license terms.

Security is part of the reason for the caution. The company says it filtered training data to keep defensive security tasks and remove material aimed at improving attacks. It also added code security checks during training, which it credits for the higher secure-coding score.

Where You Can Use It

Qwen4-Max is live on QwenCloud through a pay-as-you-go API or the Token Plan subscription. A preview had been open to subscribers since November 11, so this release mainly adds metered API access. You can also reach the model in these places:

  • Qwen Studio, the company’s chat and creation app
  • QwenWork, its workplace assistant
  • Qoder, its coding tool
  • API regions including Singapore, Frankfurt, Tokyo, Hong Kong, Beijing and US (Virginia)

The API works with both OpenAI-style and Anthropic-style request formats. That makes it easier to test inside tools you already use, much like earlier budget options such as the DeepSeek V4.1 update.

What This Means for You

For most people, the practical story is price pressure. Each strong, cheaper model from China pushes US labs to answer on cost, as you saw around the Opus 5.5 launch. If you build with AI, a few steps are worth taking:

  1. Run your own tasks before trusting benchmark claims, since most of the published numbers come from Alibaba.
  2. Check the pricing page for your region, because the quoted rates apply to Singapore and differ elsewhere.
  3. Review your data rules first, as prompts go to Alibaba’s cloud and this model cannot be self-hosted.
  4. Wait for the smaller open models if you need something you can run on your own hardware.

If you mostly use ChatGPT or a similar app, nothing changes today. Still, competition like this tends to show up later as lower prices and faster upgrades in the tools you already pay for. You can follow the next moves in our AI news section.

Sources

Primary source: Qwen’s Qwen4-Max launch post. Additional reporting from The Straits Times (Bloomberg) and The Defiant.

Conclusion

Qwen4-Max gives Alibaba a new flagship that is cheaper to call and, by the company’s own tests, better at long professional tasks than the model it replaces. The big caveats are that most of the evidence comes from Alibaba itself and that this version is hosted only, a shift for a family built on open releases. Smaller open Qwen4 models are promised but undated. Until independent results arrive, the safest move is to test it on your own work and compare the bill.

Newsletter

New AI tools in your inbox

The best new tools and the AI news that matters, in one short email. Unsubscribe any time.

Free · No spam · Unsubscribe any time

✓ Looks like you’re subscribed

Loading Google’s sign-up form…

Looks like you’re already subscribed

Thanks for reading! New AI tools and the AI news that matters will land in your inbox.

Not subscribed yet? Refresh the page to load the sign-up form.

The sign-up form couldn’t load. If you use an ad or content blocker, allow news.google.com and reload the page.

Free instant check before you pay. Takes a few seconds.

Scroll to Top