Alibaba's Qwen3.8-Max is 2.4 trillion parameters and going open-weight
Alibaba dropped Qwen3.8-Max-Preview today. 2.4 trillion parameters, fully multimodal, and the team says it will go open-weight soon. That last part matters more than the benchmark claims.
The model is live right now on Alibaba’s Token Plan, Qoder, and QoderWork platforms. The official Qwen account described it as “one of the most powerful models available today, compatible to leading frontier AI models, second only to Fable 5.” Bloomberg’s writeup frames it the same way: comparable to frontier models, trailing only Anthropic’s Fable 5.
That “second only to Fable 5” framing is doing a lot of work. Fable 5 is Anthropic’s current flagship. Being second to it, if the benchmarks hold up under independent testing, would put Qwen3.8-Max ahead of GPT-6, Gemini 3.5 Pro, and Grok 2T. That is a bold claim from a lab that was written off as a follower eighteen months ago.
The technical details are thin so far. 2.4T parameters, fully multimodal, sharper on code engineering and professional cowork. Developers with early access report very strong frontend code generation. No architecture paper, no context window spec, no MoE routing details yet. This is a preview in the literal sense: Alibaba wants the benchmark headlines before the technical report.
What makes this more than another frontier model launch is the open-weight commitment. Qwen’s official account confirmed the full Qwen3.8 release will be open-weighted. If that happens, it will be the second multi-trillion-parameter open-weight model this year, alongside Moonshot AI’s Kimi K3 at 2.8T. Two open models at this scale changes the economics of the entire industry.
The FirstPost reporting from this morning lays out why Silicon Valley is paying attention. On OpenRouter, the developer marketplace where most production AI traffic now flows, Chinese models occupy the top five positions by weekly token usage. Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai. They are all open-weight. Industry estimates suggest open-weight models can already handle around 95 percent of enterprise AI queries, leaving only the hardest tasks for closed frontier models from OpenAI and Anthropic.
Alibaba has been the most prolific open-weight lab in the world by volume. Over 400 Qwen variants, more than a billion downloads, the most-downloaded open model family globally. But the flagship Max series has stayed closed and API-only since Qwen3.7-Max launched in May. Breaking that pattern with Qwen3.8 would be a real shift.
The pricing play is also new. Alibaba launched Token Plan alongside the preview, a personal subscription that covers multiple flagship models. Team pricing dropped at the same time. This is Alibaba trying to compete with ChatGPT Plus and Claude Pro directly, not just sell API access to developers.
The competitive pressure on closed labs is now obvious. OpenAI is preparing a public offering. Anthropic is preparing a public offering. Both are reportedly targeting valuations approaching a trillion dollars. Meanwhile, Chinese labs are giving away models at the same capability tier and promising open weights within weeks of launch. You can either pay for the API or download the model, fine-tune it, and run it on your own GPUs.
The technical report will tell us whether the benchmark claims hold. The open-weight release will tell us whether the model is actually as good as advertised. Until then, Qwen3.8-Max-Preview is a very loud shot across the bow.