2 min read
The great Chinese model firehose kept blasting this week. On 19 July, Alibaba previewed Qwen3.8-Max, a 2.4-trillion-parameter multimodal monster it cheerfully bills as one of the most powerful models on Earth, second only to Anthropic’s Fable 5. That is a genuinely enormous claim. It is also, as of publication, backed by precisely nothing you could check.
What Alibaba actually put out
A preview. Qwen3.8-Max-Preview is live on Alibaba Cloud’s coding platforms, it takes images, video and documents as input, and it is the team’s first multimodal model above a trillion parameters. The full Qwen3.8 is promised as open-weight, which is the genuinely interesting part, since it would hand a frontier-class model to anyone with the hardware to run it.
Here is what does not exist yet: a model card, a verifiable parameter count, a context window, a benchmark table, per-token pricing, a licence, or an actual launch post from Qwen. “Second-best model in the world” is currently a sentence in a marketing preview, not a measured result. In fairness, that is how a great many of these launches now work: announce the vibe, produce the receipts later, ride the headlines in between.
The part that should worry the US labs
What should trouble the American labs is the trend, not the single announcement. Qwen3.8-Max landed just two days after Moonshot AI’s Kimi K3, a 2.8-trillion-parameter open-weight model, and hot on the heels of Z.AI’s GLM-5.2. China is not releasing one big model, it is releasing a big model roughly every time you refill your coffee, and increasingly it is giving the weights away for nothing. That combination, frontier-scale ambition plus open weights, is exactly the pincer movement that erodes the pricing power of the closed American labs. You cannot charge a premium for a capability your rival is handing out for free.
Whether Qwen3.8-Max really is the second-best model alive, we will find out when someone independent finally gets to poke it. Until then, treat the ranking the way you would treat any number a company invents about its own product: politely, and with your wallet firmly shut.
The AI tool stack actually worth paying for
One email a week. The tools, models and moves that matter, minus the hype and the horseshit filter set to maximum. Free.
Did you know: Alibaba’s Qwen family is already one of the most downloaded open model lineages on the planet, with tens of thousands of derivative models built on top of it. The open-weight strategy is not charity, it is distribution: get the whole world building on your architecture, and you quietly set the standard everyone else has to follow.