Alibaba’s Qwen team confirmed that its next flagship system, Qwen3.8, is headed for an open-weight release built around 2.4 trillion parameters. A preview build, Qwen3.8-Max-Preview, is already live through Alibaba’s Token Plan, plus the company’s Qoder and QoderWork products, ahead of the full weights drop.
The timing is the part worth noticing. Qwen3.8 arrives only days after Moonshot AI, the Beijing-based lab behind the Kimi model line, released Kimi K3 with 2.8 trillion parameters under an open license. Two Chinese labs have now put multi-trillion-parameter systems into developers’ hands inside the same short window, a cadence no US frontier lab has matched at comparable scale or licensing terms this year.
That cadence, not the raw parameter count, is the actual news. Anthropic, OpenAI, and Google have kept their largest, most capable systems closed, distributing access through metered APIs rather than downloadable weights. Alibaba and Moonshot are running the opposite bet: give away frontier-scale weights and monetize the infrastructure around them, in Alibaba’s case through Token Plan pricing tied to Qoder and QoderWork usage.
Qwen’s own post made a bold claim about where the model stands. It described Qwen3.8 as among the most capable systems available today and said it trails only one named rival, Fable 5, a comparison the announcement did not back with any benchmark scores. No independent evaluation results accompanied the release, so the claim of near-frontier standing currently rests entirely on Qwen’s own description of its model.
Open-weight releases at this scale change the competitive math differently than closed launches do. Once weights are public, any team with sufficient compute can fine-tune, distill, or redeploy the model without a commercial relationship with Alibaba. A fine-tuned derivative built on a free 2.4 trillion parameter base costs a fraction of an equivalent closed-model API contract at production scale, a gap that compounds the more usage a team runs through it.
For closed US labs, the standard response has been to hold their release policy while pointing to frontier capability and safety review as the differentiator worth paying for. Two trillion-parameter-class open releases from Chinese labs inside one week narrows that argument to whatever capability gap remains once both models are actually benchmarked, since neither Alibaba nor Moonshot has published comparative scores against Western frontier systems yet.
Qwen3.8-Max-Preview is available now for teams willing to run an unreleased model against real workloads ahead of the open-weight drop. Operators evaluating Chinese open-weight options for cost-sensitive deployments should treat Qwen3.8 and Kimi K3 as a pair to benchmark against each other, and against their current closed-model vendor, before locking in a 2026 infrastructure contract.
Alibaba’s Qwen team announced Qwen3.8 in a post on its official X account, @Alibaba_Qwen, a day before this article’s publication.
