⚡ Quick verdict

We may earn a commission if you buy through our links, at no extra cost to you.

Alibaba launched Qwen4-Max on 16 November 2026, and the launch was designed to make noise: a 2.4-trillion-parameter mixture-of-experts model, a headline coding score of 75.7 on DeepSWE v1.1, and claims of beating the best models from Anthropic and Moonshot. A preview had been running quietly since 11 November, so the launch was less a surprise than a confirmation — but the details matter more than the drumbeat, because this is a very different kind of Qwen release.

The Qwen family built its reputation on open weights. Developers could download the models, run them locally, and fine-tune them freely, and that openness is a big reason Qwen became one of the most-used model families in the world. Qwen4-Max breaks that pattern: it is closed, available only through Alibaba's API, with smaller Qwen4 models promised later. That single decision changes who this model is for — and it is the first thing to understand before the benchmarks.

What is Qwen4-Max?

Qwen4-Max is Alibaba's new flagship large language model, the top tier of the Qwen4 family. At 2.4 trillion total parameters built as a mixture-of-experts (MoE) architecture, it is one of the largest models Alibaba has shipped — but the parameter count alone tells you little. In an MoE design only a fraction of those parameters are active for any given token, so the meaningful questions are what it can do and what it costs to run, not how big the headline number is.

Alibaba positions Qwen4-Max squarely at coding and agentic work: long software-engineering tasks, multi-step tool use, and autonomous agents that plan, execute, and correct themselves. The DeepSWE v1.1 score of 75.7 is the centerpiece of that pitch. DeepSWE tests models on real software-engineering problems — the kind that take a human developer hours, not seconds — so a score in the mid-70s is a genuinely strong result that places Qwen4-Max among the best coding models available today. For the broader coding-model landscape, see our best AI coding models of 2026 roundup.

What Alibaba has not yet confirmed in the launch materials we reviewed is a context-window figure — check the official Qwen documentation for the current number before planning long-context workloads. Availability details are also still settling: the model is live through Alibaba Cloud's API, and whether and when it appears in consumer-facing Qwen chat products is worth confirming on the official site rather than assuming.

The benchmarks: strong, but read the fine print

Here is where an honest review has to slow down. Launch-day benchmark claims are always vendor-selected, and Qwen4-Max's launch is a textbook case of why that matters.

First, Alibaba's own comparison chart shows Qwen4-Max leading on only 4 of 16 benchmarks. That is not a bad result — leading on a quarter of a broad benchmark suite, including the coding tests that matter most for its target users, is credible. But it is not the sweep the headlines imply, and the other 12 benchmarks are where competitors still lead. If your workload is general reasoning, math, or long-form writing rather than code, the headline score may not describe your experience at all.

Second, the reporting itself is confused. Bloomberg's coverage says Qwen4-Max beat Claude Fable 5.1 and Kimi K3.1 Pro; other outlets reporting the same launch name Claude Opus 5.5 instead. Two different Claude model names for the same comparison, published on the same day, is a red flag — it suggests at least one report is working from garbled or outdated comparison tables. We attribute every head-to-head claim in this article to Alibaba's own figures, and we recommend waiting for independent benchmarks before anyone declares a winner. The honest position on 16 November is: Qwen4-Max looks like a top-tier coding model on Alibaba's numbers, and the independent picture is still forming.

For context on how these flagship comparisons usually play out, our Gemini 4 Hydrogen vs Claude Fable 5.2 comparison shows how close — and how workload-dependent — the top of the market has become.

API pricing: flagship money, fairly charged

Qwen4-Max costs $1.50 per million input tokens and $4.50 per million output tokens through Alibaba's API. That is firmly in flagship territory: more expensive than the budget and mid-tier models that cost pennies per million tokens, but competitive with other top-tier models, several of which charge more for output tokens.

The pricing structure tells you who Alibaba is targeting. Input tokens are cheap relative to output, which favors agentic workloads where the model reads a lot of context — a large codebase, long documents, tool outputs — and writes relatively compact responses. If your use case is the reverse (short prompts, long generations), the $4.50 output rate is the number that will dominate your bill. As always, check the official Alibaba Cloud pricing page before budgeting: launch pricing has a habit of changing in the first weeks.

There is no free tier for Qwen4-Max that we could confirm at launch — check the official site for any trial credits or promotional allowances, which are common around a flagship release and easy to miss.

Who Qwen4-Max is actually for

It is not, at launch, for developers who need open weights, local deployment, or fine-tuning on their own hardware. For those users the right move is to wait for the smaller Qwen4 models Alibaba has promised — those are the releases most likely to continue the family's open tradition.

What to watch before you commit

Three things will decide whether Qwen4-Max is a genuine frontier model or a well-marketed launch. First, independent benchmarks: the next few weeks of third-party evaluations will either confirm the DeepSWE result or put it in context. Second, the smaller Qwen4 releases: their openness, size, and performance will show whether Alibaba is keeping its open-weight commitment at the tiers most developers actually use. Third, real-world agentic reliability — how often the model completes multi-step tasks without drifting, hallucinating tool calls, or needing human rescue. That is the metric no launch chart captures, and it is the one that matters most for the audience Alibaba is courting.

The verdict

Qwen4-Max is a serious model with a serious coding result, launched with the usual amount of vendor spin. The DeepSWE 75.7 score is worth taking seriously; the "beats Claude" headlines are worth taking with salt until independent tests land; and the closed-weights decision is the real story of this release — it tells you Alibaba now sees its flagship as a cloud product, not a community artifact. Try it if you build coding agents and can live with an API-only model. Wait if you need openness. And if your current model already does the job, there is no reason to switch on launch-day claims alone.

Frequently asked questions

What is Qwen4-Max?

Qwen4-Max is Alibaba's new flagship AI model, launched on 16 November 2026. It is a 2.4-trillion-parameter mixture-of-experts (MoE) model positioned as Alibaba's top-tier offering for coding and agentic work, and it is available through Alibaba's API rather than as an open-weight download.

Is Qwen4-Max open source?

No — not at launch. Qwen4-Max is a closed, API-only model. That is a break from the Qwen family's open-weight tradition. Alibaba has said smaller Qwen4 models are coming later, and those are the ones most likely to be released openly, but check the official Qwen pages for the latest status before assuming anything is downloadable.

How much does Qwen4-Max cost?

At launch, API pricing is $1.50 per million input tokens and $4.50 per million output tokens. That puts it in the same price band as other flagship models — cheaper than some, more expensive than budget tiers. Prices can change, so check the official Alibaba Cloud pricing page before budgeting a project.

How good is Qwen4-Max at coding?

On Alibaba's own figures, Qwen4-Max scored 75.7 on DeepSWE v1.1, a demanding software-engineering benchmark — a genuinely strong result that puts it in the top tier for coding. But benchmarks are vendor-selected: Alibaba's own comparison chart shows Qwen4-Max leading on only 4 of 16 benchmarks, so treat the headline number as one data point, not the whole story.

Did Qwen4-Max beat Claude and Kimi?

It depends on which report you read, which is exactly why you should be careful. Bloomberg's coverage says Qwen4-Max beat Claude Fable 5.1 and Kimi K3.1 Pro on Alibaba's chosen benchmarks; other outlets name Claude Opus 5.5 instead. The model names themselves differ between reports published on the same day, so we attribute all head-to-head claims to Alibaba's own figures and recommend waiting for independent benchmarks before declaring a winner.

Should I switch to Qwen4-Max?

If you build coding agents or run heavy agentic workloads through an API, Qwen4-Max is worth a trial — the DeepSWE score and the $1.50/$4.50 pricing are competitive. If you need open weights, local deployment, or a model you can fine-tune yourself, wait for the smaller Qwen4 releases. And if your current setup already works, there is no urgent reason to switch on launch-day claims alone.

AT
About Aitopkit

We research AI tools in depth before recommending them — official pricing, public docs, and real user reports. If you buy through our links we may earn a commission at no extra cost to you. Read our disclosure.