Alibaba previewed a 2.4 trillion-parameter artificial intelligence model on Sunday that the company says ranks behind only one system in the world today: Anthropic’s Claude Fable 5. Hong Kong-listed Alibaba shares jumped as much as 5.4 per cent on Monday on the news, even as the company offered no benchmarks, model card, or independent test to back the claim. (Source: straitstimes.com)
What Qwen3.8 Max Preview Actually Is
Qwen3.8 Max Preview is the Hangzhou-based cloud giant’s largest model to date and the first in the Qwen family above a trillion parameters that is fully multimodal, handling images, video, and documents alongside text. The model is now live on Alibaba’s Token Plan subscription and on its Qoder and QoderWork coding platforms, priced at roughly 10 per cent of the standard rate during the trial window. Open weights are promised but the date has not been set and no licence terms have been disclosed. (Source: officchai.com)
The parameter count is the headline. Two trillion-parameter-plus systems were essentially unheard of outside China’s big-tech cloud labs until Moonshot AI shipped Kimi K3 at 2.8 trillion parameters days earlier. Alibaba’s 2.4 trillion-parameter Qwen3.8 Max sits in roughly the same weight class, joining K3 in a new bracket that puts Chinese open-weight efforts within striking distance of the largest US frontier systems. (Source: thenextweb.com)
The ‘Second Only to Fable 5’ Claim, Explained
Alibaba’s Qwen account on X posted that Qwen3.8 is “one of the most powerful models available today, compatible to leading frontier AI models, second only to Fable 5.” That puts Anthropic’s Mythos-class model at the top of the company’s internal ranking and Qwen3.8 in second place. The framing matters because Alibaba owns roughly 36 per cent of Moonshot AI, the lab behind Kimi K3, making the launch as much an in-house rivalry as a competitive shot at Anthropic. (Source: officchai.com)
The catch is the receipts. SiliconANGLE reported that Alibaba’s flagship landed without the published benchmark suite that has accompanied every previous Qwen flagship, including Qwen3.7 Max earlier this year. On the public LMArena leaderboard, the older Qwen3.7 Max ranks well below Fable 5, and no public score has yet been published for Qwen3.8. Until weights drop, the “second only” framing is Alibaba’s own assessment rather than an independently verified position. (Source: thenextweb.com)
The Bigger Signal: China No Longer Treats Open-Weight as a Consolation Prize
Qwen3.8 arrives three days after Moonshot AI’s Kimi K3 release and one day after Bloomberg confirmed China’s Z.AI shipped GLM-5.2 within striking distance of Claude Opus 4.8 on the FrontierSWE coding benchmark. Three Chinese labs in roughly a week have produced trillion-parameter-plus open-weight systems pitched as competitive with the best closed US models. (Source: livemint.com)
That is a meaningful break from the playbook of 2024 and early 2025, when open-weight Chinese releases were typically framed as 70 to 80 per cent of US frontier capability. The new releases pitch themselves as direct competitors, and Chinese e-commerce and financial markets have responded. Z.AI fell 14 per cent on Monday after a 30 per cent slide on Thursday, while Alibaba extended a months-long rally. Moonshot’s Kimi K3 was so heavily subscribed that the company paused new sign-ups on Sunday night. (Source: straitstimes.com)
What It Means for Anthropic, OpenAI, and the Wider US AI Stack
Anthropic’s revenue model rests on selling paid API access to large frontier models that open-weight alternatives are catching up to on raw benchmarks and starting to match on coding, the highest-margin workload in AI. If the open-weight releases from Moonshot and Alibaba can credibly serve enterprise coding use cases through inference providers like Alibaba Cloud, Together AI, or open-source self-hosted clusters, the incentive for enterprises to keep paying Anthropic premium rates narrows. (Source: livemint.com)
“Competition is likely to intensify, eventually leaving only a small group of large language model developers capable of funding frontier-model development,” said Laila Khawaja, research director at Gavekal Technologies. (Source: livemint.com)
The Open-Weight Strategy Behind the Headlines
Alibaba’s launch structure is telling. The company is putting Qwen3.8 into paid products first and promising open weights later, the same pattern Alibaba used when scaling its previous trillion-parameter Qwen3-Max-Preview back in September 2025. The two-step pattern turns launch buzz into paying customers before allowing outside researchers to inspect the weights or run local evals. (Source: officchai.com)
AI researcher Nathan Lambert said “something is changing” about the willingness of China’s largest labs to ship their top models as open weights, framing the Qwen3.8 release as the start of a more aggressive phase. The weights, once released, would let any researcher reproduce Alibaba’s ranking claim. For now, the company’s stock is moving on the announcement, the parameters, and the perceived competitive position, not the public benchmark scorecard. (Source: thenextweb.com)
Alibaba’s Qwen3.8 Max is “one of the most powerful models available today, compatible to leading frontier AI models, second only to Fable 5.”
What To Watch This Week
Three dates will determine whether Qwen3.8 Max becomes a real Fable 5 challenger or a marketing-grade announcement. The first is the public release of open weights, which Alibaba has pledged without a date. The second is independent evaluation on LMArena, the live human preference leaderboard. The third is Moonshot AI’s planned IPO within the next six months, which would mark the first Chinese open-weight trillion-parameter lab to list publicly and reset the funding benchmark for every comparable model. (Source: livemint.com)
For Anthropic, the operating question is whether closed frontier models can hold a pricing premium over increasingly competitive open-weight alternatives as more trillion-parameter-plus systems ship from Chinese labs in the second half of 2026.