Fable 5 on Benchmarks, and on the Arena Text Leaderboard, Qwen3.8-max Trails Only Fable 5 and Three Opus Models

Alibaba's own testing shows performance broadly matching or exceeding Fable 5 on benchmarks, and on the Arena text leaderboard, Qwen3.8-Max trails only Fable 5 and three Opus models. For frontend coding, it is beaten only by two Claude Opus models and Kimi K3, while only Fable 5 outperforms it in visual analysis. The model has 2.4 trillion parameters, compared with Kimi K3's 2.8 trillion, though most top US labs keep such figures private.

Alibaba plans to release Qwen3.8-Max's weights next week, marking a return to open-weight releases after a brief pivot to proprietary models earlier this year. The move intensifies competition with US firms, as Chinese companies accelerate releases, including Kimi K3 and new video generation models from ByteDance and MiniMax. Openness remains divisive in the US, where industry leaders largely support open-weight models for safety and competition, even as closed-model providers face scrutiny over cyberattacks by their own AI agents.