Key Takeaways:
- Alibaba launched Qwen3.8-Max-Preview, a 2.4 trillion-parameter AI model
- The company claims it trails only Anthropic's Fable 5 in benchmark performance
- Token Plan Personal Edition offers tiered pricing from $5.40 to $69.30 per month
Quay lại
Alibaba launches 2.4 trillion-parameter Qwen3.8, trailing only Fable 5
Three Chinese AI labs have released frontier-class models in as many months, and Alibaba's Qwen3.8-Max-Preview — a 2.4 trillion-parameter system the company says is second only to Anthropic's Fable 5 — is the latest bid to prove open-weight models can compete with the best behind closed APIs.
Alibaba Group Holding Ltd. on Sunday launched Qwen3.8-Max-Preview, a 2.4 trillion-parameter large language model that the company claims trails only Claude Fable 5 among models it benchmarked against. The model debuted on Alibaba's newly released Token Plan Personal Edition, a subscription service offering tiered access to the company's AI infrastructure, with open weights promised for a future release.
"Qwen3.8 is one of the most powerful models available today, comparable to leading frontier AI models, second only to Fable 5," the Qwen team said in a post on X. The company did not publish independent third-party benchmark results, leaving the claim pending verification from outlets such as Artificial Analysis and LMArena.
The 2.4 trillion-parameter count marks a significant jump from Alibaba's previous flagship. Qwen3-Max-Preview, launched in September 2025, was the company's first model to cross the trillion-parameter threshold. Qwen 3.6 Max Preview followed in April 2026. The new model now sits at a similar scale to Moonshot AI's Kimi K3, which launched at 2.8 trillion parameters and briefly held the title of the largest open-weight model ever released. Alibaba owns roughly 36% of Moonshot AI, making the two releases an in-house rivalry as much as a competitive one.
The model excels in code engineering and professional office scenarios, according to Alibaba, and is available immediately through the company's Qoder and QoderWork coding platforms, as well as the Qwen PC client for free trial use.
Token Plan Pricing and Strategy
Alibaba's Token Plan Personal Edition offers three tiers: Lite at 39 yuan ($5.40) per month, Standard at 139 yuan ($19.30) per month, and Pro at 499 yuan ($69.30) per month, with introductory discounts ranging from 6.5折 to 8.3折 off regular pricing. Team plans are also available, starting at 150 yuan per seat per month for a standard tier up to 1,398 yuan per seat per month for a premium tier.
As a launch promotion, daytime credit consumption is discounted to 10% of normal rates, and personal Token Plan users receive an additional 80% discount on nighttime usage. The plans support compatibility with tools built around OpenAI and Anthropic's protocols, allowing developers using existing coding agents to point them at Qwen3.8 without rebuilding workflows.
The Chinese AI Offensive
Qwen3.8 joins a wave of Chinese open-weight releases that have reshaped the global AI competitive landscape. GLM 5.2 from Z.AI landed within striking distance of Claude Opus 4.8 on FrontierSWE and drew praise from executives at Vercel and Box. Kimi K3 went further, beating both Claude Fable 5 and GPT-5.6 Sol outright on Arena's Frontend Code leaderboard, becoming the first open-weight model to top that ranking.
The pattern marks a departure from the earlier narrative that Chinese labs were trailing Western frontier developers. Three major Chinese releases in three months — GLM 5.2, Kimi K3, and now Qwen3.8 — each claim a spot at or near the top of global leaderboards, and each is either already open-weight or headed there.
Investment Implications
Alibaba's aggressive push into frontier AI comes as the company's cloud division competes for enterprise AI workloads against Baidu, Tencent, and global providers such as Amazon Web Services and Microsoft Azure. The Token Plan represents a monetization strategy for Alibaba's AI infrastructure, offering a recurring revenue stream tied to model access rather than one-off API calls.
The open-weight commitment, while delayed relative to previous releases, positions Alibaba to capture developer mindshare in a market where Meta's Llama series and Mistral have set expectations for freely available frontier models. If Qwen3.8's performance claims hold up under independent testing, Alibaba could challenge the assumption that the best AI models require proprietary APIs — a dynamic that would pressure inference pricing across the industry.
Alibaba's shares trade on the New York Stock Exchange under the ticker BABA. The company reports fiscal first-quarter earnings in August.
This article is for informational purposes only and does not constitute investment advice.
