Subscribe
AI Models

Alibaba unveils flagship Qwen4-Max, claiming top coding benchmarks

Monday, November 16, 2026 · 6 min read

Alibaba has unveiled Qwen4-Max, its new flagship AI model with 2.4 trillion parameters, claiming top scores on coding tests ahead of Anthropic and Moonshot. The benchmark claims are Alibaba's own.

Bullish$BABA

📊 Scorecard: our Bullish call on $BABA is scored from $94.38, the last close before it ran. Its first result comes after the next close.

Alibaba unveils flagship Qwen4-Max, claiming top coding benchmarks

The News

Alibaba has unveiled Qwen4-Max, its new flagship artificial intelligence model, with 2.4 trillion parameters. The company announced it on Monday, November 16 (Beijing time). A parameter is a number the model adjusts as it learns from data, and more parameters generally mean a bigger, more capable model, though size alone does not decide quality. At 2.4 trillion, Qwen4-Max is one of the largest models any company has publicly described.

Bloomberg reported that Alibaba says the model scored near the top of coding benchmarks, outperforming Anthropic's Claude Fable 5.1 and Moonshot's Kimi K3.1 Pro, with top rankings in coding, agentic tasks and visual reasoning. Agentic tasks are jobs where the model acts on its own to complete multi-step work, such as booking a trip or fixing a piece of software, rather than just answering a single question. Visual reasoning means understanding images, charts and diagrams. These are Alibaba's own benchmark claims, not independently verified results, and benchmark scores depend heavily on which tests a company chooses to run.

The model itself is closed-weight, meaning Alibaba is not sharing the trained model files publicly, but the company says smaller Qwen4 models will be open-weight soon. Open-weight models can be downloaded and run by anyone, which has been a big part of Alibaba's strategy with its earlier Qwen models. Alibaba also said training included security checks alongside the coding tests, with safeguards against misuse in cybercrime and hazardous materials strengthened, and its internal review finding its cyber-defence capabilities improved substantially and protections against offensive use significantly reinforced.

Alibaba says Qwen4-Max scored near the top of coding benchmarks.

Why It Matters

Coding has become the most important test for AI models, because it is the job customers will pay for. A model that writes, tests and fixes software saves companies real money, and businesses are already spending heavily on coding assistants. That is why every major AI launch now leads with coding scores, and why Alibaba is measuring Qwen4-Max against Anthropic and Moonshot, the two rivals best known for coding ability.

The launch also shows how the AI race has split into two strategies. American companies such as Anthropic and OpenAI mostly keep their best models closed, charging for access. Alibaba has mixed the two, giving away smaller models as open-weight software while selling its flagship through the cloud. Open-weight models build a community of developers who learn Alibaba's tools, and that community then becomes a customer base for the paid products. Releasing smaller Qwen4 models soon keeps that flywheel spinning.

The safety notes in the announcement matter too. Alibaba said training included security checks, with safeguards against misuse in cybercrime and hazardous materials strengthened, and internal testing found no significant high-risk capabilities. As models get bigger and more capable, governments are watching more closely, and a company that can show it tested for danger before release is in a stronger position. The claims are Alibaba's own, and regulators will want proof, but the fact that the company led with them shows safety has become part of the sales pitch, not an afterthought.

Analyst's Take

Our call is Bullish. A flagship model that tops coding benchmarks, even on the maker's own tests, puts Alibaba back in the front rank of the AI race, and coding is the skill that matters most to paying customers right now. Companies buy AI to write and fix software, and a model that beats Anthropic and Moonshot at that job will get attention from developers. The promise of smaller open-weight Qwen4 models soon is the other half of the story: Alibaba keeps its huge community of developers while selling its biggest model through its cloud.

The claim should still be read with care. Alibaba chose the benchmarks, ran the tests, and reported the results, and rival companies will run their own tests and tell a different story within days. That is normal in the AI industry, where every launch comes with a chart showing the new model on top. What matters is whether independent testers and real developers agree once they get their hands on it. If they do, Qwen4-Max is a genuine step forward. If they do not, it is a good press release.

For Alibaba, the model is also a cloud story. The company makes its money in AI by renting computing power and selling model access through Alibaba Cloud, and a stronger flagship gives it a better product to sell. The share price barely moved on Friday, up 0.02% to $94.38, because the launch came over the weekend and the market will judge it on evidence, not announcements. Watch for two things: independent benchmark results in the coming weeks, and the release of the smaller open-weight models, which will show whether Alibaba is serious about keeping developers on its side. Reader poll: Bullish or bearish on $BABA? Vote: Bullish, Bearish, Mixed, Too early to tell.

Are you bullish or bearish on this story?

By The Numbers

2.4 trillionParameters in Qwen4-Max
$94.38Alibaba share price, Fri Nov. 13
+0.02%Alibaba share move, Fri Nov. 13
3Benchmark areas with top rankings: coding, agentic tasks, visual reasoning
2Rival models Alibaba says Qwen4-Max outperformed
NoneSignificant high-risk capabilities found in internal testing

FAQ

How big is Qwen4-Max?

It has 2.4 trillion parameters, the numbers a model adjusts as it learns. More parameters generally mean a bigger, more capable model, though size alone does not decide quality.

Is Qwen4-Max open-weight?

No. The flagship model is closed-weight, meaning Alibaba is not sharing the trained model files publicly. The company says smaller Qwen4 models will be open-weight soon.

Are the benchmark claims independently verified?

No. The benchmark results are Alibaba's own claims, not independently verified results. The company says Qwen4-Max outperformed Anthropic's Claude Fable 5.1 and Moonshot's Kimi K3.1 Pro on coding tests.

👁

Source: Morningstar (Dow Jones)

Get the daily briefing. AI news with a markets lens — tickers, bullish/bearish reads, zero hype. Subscribe free on Substack

More on $BABA

All $BABA coverage →

More from AI Agent Update