Alibaba's Qwen3.8 Max Preview Debuts at WAIC, Claims Second Only to Anthropic's Flagship
Claire Weston
Alibaba unveiled the 2.4-trillion-parameter Qwen3.8 Max preview at Shanghai's WAIC, calling it second only to Anthropic's flagship Fable 5 — Chinese AI labs have entered the multi-trillion-parameter race, and commercial pricing dropped the same day.
How big is this model, really?
Qwen3.8 Max packs 2.4 trillion parameters. Alibaba's official post says it "rivals frontier AI models, second only to Fable 5."
This means → Alibaba is benchmarking itself directly against Anthropic's flagship, skipping the domestic mid-tier entirely.
The "second only to" claim is Alibaba's own; no independent benchmark has confirmed it yet — read the positioning with a grain of caution.
How does it stack up against Moonshot's Kimi K3?
Moonshot AI (月之暗面) released Kimi K3 last Thursday with 2.8 trillion parameters — currently considered the world's largest open-source AI model.
In plain terms = by raw parameter count, Kimi K3 is 400 billion parameters larger; but more parameters do not guarantee better performance — real-world benchmarks will decide.
The two launches came just days apart. This reflects a broader shift: Chinese AI developers have collectively entered the "multi-trillion-parameter" era, and the arms race is accelerating.
How can users access it, and what does it cost?
The preview is live now. Developers can access it via Alibaba's Token Plan subscription, plus the coding platforms Qoder and QoderWork.
Personal tiers come in three levels: Lite at ¥39/month, Standard at ¥139/month, Pro at ¥499/month — all under a limited-time discount.
Daytime credit consumption is billed at 10% of list price; personal-plan users get an additional 80% off at night. This means → Alibaba is using aggressive pricing to grab early adopters, prioritising ecosystem scale over short-term revenue.
Why does the Apple partnership matter?
Alibaba recently won Beijing's approval to serve as Apple Intelligence's technology partner in China.
This means → Alibaba's models won't just run on its own platforms — they could be embedded in the daily workflow of hundreds of millions of iPhones, a commercial surface far larger than API subscriptions alone.
In plain terms = a model is just a lab product without real-world distribution; the Apple deal moves Alibaba one step closer to turning model capability into actual revenue.
What should we watch next?
Alibaba says the full release will come soon with open weights, giving outside developers and researchers a chance to run independent benchmarks.
Two key questions remain: first, whether real-world test results sustain the "second only to Fable 5" claim; second, whether enterprise clients are willing to migrate from existing solutions.
This reflects a competition that has moved beyond parameter counts into a three-front race: scale + commercialisation + ecosystem lock-in — size alone no longer wins.
Content is for reference only, not financial advice.