Alibaba’s flagship Qwen4-Max went on general sale overnight at a quarter less than its predecessor — and Tencent answered the same day with a model 21 times cheaper still. The AI race is now a price war, fought on the API meter.
UPDATE: Qwen4-Max goes live — Alibaba’s flagship, 25% cheaper than the model it replaces
Five days after its preview opened, Alibaba has moved Qwen4-Max to general availability. The model went live on November 16 on the QwenCloud API, in Qwen Studio, QwenWork and Qoder, and within hours on OpenRouter and OpenCode Go. The preview identifiers retire on November 25; the live model ID is simply qwen4-max.
Alibaba’s launch post fills in what the preview left out. Qwen4-Max is a mixture-of-experts model with a 2.4-trillion-parameter backbone and 71 billion parameters active per token, with native vision: it takes text, images and video as input, with a 1-million-token context window. In thinking mode, reasoning and the final answer share a 262,144-token budget. Alibaba pitches it as a teammate for coding, professional work and long-running agents.
The number that matters most is the price. QwenCloud charges $1.50 per million input tokens and $4.50 per million output tokens, with cached input at $0.075 — a straight 25% cut on both sides of the meter against Qwen3.8-Max at $2.00 and $6.00, and 70% cheaper on cached input. The model is also on OpenRouter at the same rates.
The benchmarks deserve the usual caution. Alibaba’s own scorecard claims gains on professional work, but the independent Artificial Analysis board lists Qwen4-Max at 43 on its Intelligence Index, behind Moonshot’s Kimi K3.1 Pro at 46 and well behind Claude Fable 5.2 at 65. This is a cheaper flagship, not a new leader — and unlike Qwen’s open-weight releases, Qwen4-Max ships as an API only, with no downloadable weights.
Why it matters: the price cut is the story. Alibaba is betting that volume beats margin, and that a flagship priced below its own predecessor pulls developers off rival APIs. For Canadian developers, a frontier-class model at $1.50 per million input tokens lowers the cost of building AI features — but the closed, API-only release also means the model can only be rented, never run locally. (enjoy.dev, OrcaRouter, ApiDog, AIstify)
Tencent answers the same day with Hy4 flash — 21 times cheaper on input
Tencent’s Hunyuan team launched Hy4 flash on November 16, the same day Qwen4-Max went live. The model has 224 billion total parameters with 12 billion active per token, a 1-million-token context window and multimodal input — and a price card that undercuts Alibaba’s flagship by an order of magnitude: $0.070 per million input tokens and $0.281 output, with cached input at $0.003. That is roughly 21 times cheaper on input than Qwen4-Max, and about 5% of the cost of Tencent’s own full-size Hy4, which launched on November 4.
Why it matters: the two launches describe a two-tier market — Alibaba competing for heavy agentic work, Tencent pricing for volume. Flagships are getting cheaper, and small models are getting cheaper faster. (OrcaRouter)
Pony.ai’s CFO says China’s self-driving firms have practised their Nvidia Plan B for years
Chinese self-driving companies have spent years building a non-Nvidia chip backup plan, Pony.ai chief financial officer Leo Wang told TechCrunch Mobility in an interview at Web Summit in Lisbon, published November 15. “Everyone, I would say, is also practicing kind of like a Plan B for a few years, so at least if you think about that, I would say there are alternatives,” Wang said. “Although cost-effectiveness-wise, probably it’s not the best choice yet.”
The context is the squeeze we covered last week: demand for computing already exceeds supply, and Pony.ai is turning more to domestically produced chips for AI training while directing engineers to use mid-tier compute where possible. Wang was asked what happens if the fallout from Jensen Huang’s resignation leads to even stricter export controls on Nvidia chips shipped to China. No new restriction has been announced — the trigger is hypothetical — but the Plan B he describes is about training compute, not the chips in the cars, and the cost gap against Nvidia has not closed. (TechCrunch, Superpower Daily)
Moonshot AI unwinds its Cayman structure to clear the way for a Hong Kong IPO
Moonshot AI, the Beijing maker of the Kimi models, is wrapping up a corporate restructuring to pave the way for an initial public offering, the Wall Street Journal reported on November 13, citing people familiar with the matter. Under the deal, Moonshot is deactivating its Cayman Islands holding company — where many early backers invested in dollars — and moving investors’ stakes to its main operating entity in China, buying back shares from those investors, who will then reinvest in the onshore entity.
The listing plan is ambitious: a Hong Kong listing targeted for the first quarter of 2027 that could raise up to $5 billion, after a final private round closed at a valuation of about $50 billion. Moonshot filed confidentially with the Hong Kong Stock Exchange in September, according to Reuters. The South China Morning Post described the move as a full unwind of the variable-interest-entity structure, driven by tightened China Securities Regulatory Commission rules on offshore listing vehicles. Moonshot declined to comment, and the timing and size of the listing may still change.
Why it matters: Moonshot would be the first frontier Chinese AI lab to test public markets at a $50 billion valuation — and the structure of the deal says as much as the number. An onshore listing vehicle means Beijing, not offshore investors, sets the terms. (Dealroom, AI Weekly)
Beijing maps 354 AI scenarios for transport — and counts 3.1 million petaflops of computing power
China’s Ministry of Transport announced on November 14 its first batch of 354 typical application scenario projects under the “AI Plus Transportation” initiative, covering seven sectors: autonomous driving, intelligent highways, smart shipping, facility construction and maintenance, intelligent passenger transit, freight logistics and safety supervision. The programme is structured in three tiers — core technology breakthroughs, innovative pilot demonstrations and large-scale deployment.
In parallel, China’s National Data Administration disclosed at a standards conference in Xiamen, held November 9–14, that the country’s total computing capacity has reached 3.1 million petaflops, ranking second globally. The same event launched an embodied-AI data standards drafting working group, alongside four industry standards groups covering meteorology, healthcare, electric power and commerce.
Why it matters: this is the state codifying deployment, not just funding it. A published scenario map tells robotaxi operators, port builders and highway agencies exactly where AI is sanctioned to go next — and the data-standards working group is the unglamorous plumbing that decides whether robots from different makers can share training data at all. (KIMKJ)
Toyota rewires its China supply chain to fight back in the EV race
Toyota Motor is reshaping its supply chain in China, looking beyond its traditional network of affiliated Japanese suppliers — the keiretsu — to local companies for electric vehicles, Nikkei Asia reported on November 13. The shift covers batteries, power electronics and software-defined vehicle architecture, aiming to compress development cycles and match the cost structure that has given local brands such as BYD and Geely their edge.
The numbers show how far the shift has already gone. Chinese suppliers account for 65% of the supply chain for the 2026 Bozhi 3X (bZ3X), the GAC Toyota electric SUV led by a Chinese engineering team, which has topped joint-venture EV sales for seven consecutive months. Toyota has also transferred decision-making for China-exclusive models to local teams, and plans to procure parts from Chinese manufacturers at its largest Southeast Asian base in Thailand, applying them to new EVs produced there starting in 2028.
Why it matters: the world’s biggest automaker is conceding that China’s cost and speed advantage is structural, not cyclical — and it is now exporting that advantage to its own plants abroad. (Nikkei Asia via Meridan Report, Gasgoo)
The big picture
Today’s six stories are one story told six ways: China’s technology build-out is being repriced, restructured and standardised at the same time. Alibaba and Tencent cut the price of frontier AI in a single day, Moonshot is rebuilding its corporate structure to take a $50 billion valuation public, Pony.ai is admitting its Nvidia backup costs more than the original, Beijing is writing the scenario map for AI in transport, and Toyota is rebuilding its supply chain around Chinese parts. The through-line is adaptation under constraint — cheaper, onshore, and built to work without the things China can no longer count on.
Watch next
Nvidia reports quarterly results on Tuesday, November 17 — interim chief executive Colette Kress’s first public outing in the role, and the first hard numbers since the resignation.
Sources
- enjoy.dev, November 16, 2026 — Qwen4-Max is out, 25% cheaper than Qwen3.8-Max: https://enjoy.dev/blog/qwen4-max-preview/
- OrcaRouter, November 16, 2026 — Qwen4-Max and Tencent Hy4 flash: two launches, one day: https://www.orcarouter.ai/blog/daily-ai-brief-nov-16-qwen4-max-hy4-flash
- ApiDog, November 16, 2026 — Qwen4-Max vs Qwen3.8: price, context, benchmarks and what changed: https://www.apidog.com/blog/qwen-4-max-vs-qwen-3-8/
- AIstify, November 16, 2026 — Alibaba launches Qwen4-Max to challenge Claude and Kimi: https://aistify.com/alibaba-qwen4-max-launch-benchmarks-api-pricing/
- TechCrunch, November 15, 2026 — TechCrunch Mobility: The Nvidia backup plan for China’s self-driving industry: https://techcrunch.com/2026/11/15/techcrunch-mobility-the-nvidia-backup-plan-for-chinas-self-driving-industry/
- Superpower Daily, November 15, 2026 — Pony.ai turns to Chinese chips as computing supply falls short, CFO says: https://superpowerdaily.com/posts/pony-ai-turns-to-chinese-chips-as-computing-supply-falls-short-cfo-says
- Dealroom, November 13, 2026 — Moonshot AI unwinds its Cayman structure to clear the way for a Hong Kong IPO: https://dealroom.co/news/172710-moonshot-ai-unwinds-its-cayman-structure-to-clear-the-way-for-a-hong-kon/
- AI Weekly, November 13, 2026 — Moonshot AI unwinds Cayman vehicle for $50B Hong Kong IPO: https://aiweekly.co/alerts/moonshot-ai-unwinds-cayman-vehicle-for-50b-hong-kong-ipo
- KIMKJ, November 15, 2026 — AI, Economy, and Global Conflict Daily Briefing, Sunday, November 15, 2026: https://kimkj.com/en/daily-briefing-en/?mod=document&uid=21717
- Nikkei Asia via Meridan Report, November 13, 2026 — Toyota rewires China supply chain to fight back in EV race: https://bangkok.meridanreport.com/article/toyota-rewires-china-supply-chain-to-fight-back-in-ev-race-yq9qkd?site=bangkok
- Gasgoo, November 2026 — Is the Japanese auto supply chain in China really undergoing a major shift?: https://autonews.gasgoo.com/articles/market-industry/2060190487536005121
