Qwen4-Max Is Out of Preview: Alibaba Ships a 2.4-Trillion-Parameter Closed Model With Built-In Web Search and Lower Prices

Five days after a preview with no model card and a 25 November deadline, Alibaba has shipped the real thing. Qwen4-Max went generally available on 16 November (Beijing time), with a model page, published prices and a public API ID, qwen4-max. It is a closed-weight model. The open-weight part of the Qwen4 family is still a promise.

What Alibaba announced

From the Alibaba Cloud announcement, the QwenCloud model page and TechNode’s Chinese report:

  • Architecture: a 2.4-trillion-parameter mixture-of-experts model with about 71 billion parameters active per token. That is 25% fewer active parameters than Qwen3.8-Max at the same 2.4T scale. Native vision input (text, image and video in, text out).
  • Context: 1M tokens, with up to 991K input and 393K output tokens.
  • Price (international): $1.50 per million input tokens, $4.50 per million output, and $0.075 per million for implicit cache hits. In China, input falls from ¥12 to ¥10 per million and output from ¥36 to ¥30, cuts of roughly 17%.
  • Efficiency claim: on Alibaba’s own Work360 benchmark, average cost per task drops from ¥26.58 to ¥11.16 (about 58%), helped by 35% fewer tokens at “High” reasoning effort.
  • Where it runs: the QwenCloud API and Token Plan, QwenWork, Qwen Studio, Qoder and, according to ifeng, the consumer Qwen (千问) app.
  • Built-in tools: the model page lists web_search, web_extractor, text-to-image and image-to-image search and a code interpreter on the Responses API, plus web_search and web_fetch on the Anthropic-compatible API.

Alibaba also claims sixth place on Code Arena and, per Reuters via the Economic Times, stronger safeguards on cyber and CBRN misuse. Those are vendor numbers on mostly vendor benchmarks, so I’d wait for independent evals before repeating the “beats US rivals” line.

Why an SEO should care about a coding model

The launch pitch is all about agents, coding and science. The search angle sits in that tools list. Qwen4-Max ships with server-side web search and page extraction. Any developer building on it gets retrieval by flipping a switch, without plugging in a third-party search API. That’s the same direction DeepSeek took on 13 November, and Baidu took when it cut its Search AI API to ¥4.9 per 1,000 calls. Chinese model vendors are bundling retrieval into the model.

Three practical points:

  1. Record the model ID when you test AI visibility. The Qwen app and QwenWork answers you audited last week may have been Qwen3.8-Max. From today they may not be. Note the date and model when you log citations.
  2. Cheaper long context means more of your page gets read. At $0.075 per million cached tokens, agents can afford to pull whole documents rather than snippets. Clear structure and answer-first sections matter more than keyword placement.
  3. The open-weight models are what spread. Max is closed. The “smaller-scale” Qwen4 models Alibaba says it will open-source soon are the ones third-party search and answer products in China and beyond will actually self-host. That release is the one to watch.

My take

The preview was a marketing tease. This release is properly documented, with a model card, prices and a lifecycle. I’ve added it to our Qwen model history page. The bigger story for search is the price. With the cache-hit rate this low, Alibaba is clearly targeting long-running agents that read the web repeatedly. Those agents will do much of the “searching” in Alibaba’s ecosystem, from Taobao to Amap. If your brand depends on Chinese discovery, it’s time to start testing how Qwen-powered agents read, extract and summarise your pages, not just how they rank them.

Dan Taylor is the founder and editor of International Search Insights (ISI), the publication behind ReadIsi, and a Partner and Head of Innovation (Search & AI) at SALT.agency. He has worked in search for more than a decade, in-house and agency-side, and is regarded as one of the West’s leading voices on search engines beyond Google, including Yandex, Baidu, Naver and Cốc Cốc. On ReadIsi, Dan covers the engines and AI answer platforms that English-language coverage tends to miss: Baidu and ERNIE, Yandex and Alice AI, Naver, Cốc Cốc, Shenma, Sogou, WeChat and Weibo search, plus DeepSeek and Alibaba’s Qwen. He also keeps dated records of documented changes, such as the Yandex, Baidu and Naver algorithm update histories and the DeepSeek, ERNIE and Qwen model update histories. His approach is to take learnings from global search engines and apply logic with a high evidence threshold. A claim needs a source, patterns seen in one market are treated as a hypothesis until they are checked in another, and where the evidence is thin he says so. For brands, that means advice built on what can be shown, with measurement that separates being cited from being chosen. Dan is credited with coining the term “edge SEO” and won the inaugural TechSEO Boost research competition in 2018 for his work on CDN and edge changes for SEO. In 2026 he was credited with coining “Infinite Tail” to describe how keyword research is changing with AI search. He is the author of AI Search Marketing For The Travel Sector, writes for Search Engine Journal, Search Engine Land and Travel Marketing Guy, and has advised technology, SaaS, travel and enterprise ecommerce brands, including Cloudflare, GitLab and Airbnb. He has given multiple talks and panels at the International Search Summit in Barcelona, and has also spoken at Optimization Moscow, brightonSEO, Google Search Central Live Deep Dive Asia in Bangkok, Athens SEO and the Asia SEO Virtual Summit 2026, where his talk looked at how Baidu, Naver, DeepSeek, Tencent, Cốc Cốc and Yandex built their own AI answers.