Alibaba ships Qwen3.8-Max, a 2.4-trillion-parameter model it's open-sourcing next week

Alibaba made Qwen3.8-Max generally available to global developers on August 3, 2026, following a July 19 preview at the World Artificial Intelligence Conference in Shanghai. It's a sparse mixture-of-experts model with 2.4 trillion total parameters and 95 billion active per token, a 1M-token context window, and native text, image, and video input.

Where it leads

Alibaba's published benchmark table shows 86.6 on Terminal-Bench 2.1, 67.7 on SWE-bench Pro, and 92.6 on GPQA Diamond, with the strongest gains showing up in multimodal and agentic categories rather than general reasoning. The model tops several vision benchmarks outright, including OSWorld-Verified (86.1), Parametric CAD Bench (91.5), and OmniDocBench 1.5 (92.1) — and ranks second globally on Arena.AI's multimodal leaderboard, behind a single Claude Fable 5 variant.

Alibaba is also pointing to a 10-day autonomous coding run in which the model built a GitHub project from an empty folder — dispatching its own issues, running its own tests, and merging its own pull requests without a human reviewing each step.

Open weights and pricing

The notable business decision here is distribution: Alibaba says open weights are coming within the week, which would make this the first time a Qwen-Max-class model has been released publicly rather than kept behind an API. A smaller sibling, Qwen3.8-27B, is being open-sourced alongside it. API pricing lands at $2 input / $6 output / $0.25 cached input per million tokens — aggressive relative to comparable Western frontier models, which is likely to keep pressure on pricing across the board through the rest of the year.

Source: MarkTechPost, Bloomberg, Neowin.