发布当日，海外主流AI平台纷纷接入阿里Qwen3.8

TL;DR - Alibaba released Qwen3.8, a 2.4T-parameter (95B active) MoE foundation model, and on launch day a wave of overseas API aggregators, agent/coding tools, and developer platforms (OpenRouter, OpenCode, Hermes Agent, Command Code, Vercel, Novita, Charm, DeepInfra) added Qwen3.8-Max support. It signals that Chinese open-weight models are becoming default options in the global developer tooling stack.

• Architecture/capability claims: 2.4T total params with 95B activated, 1M-token context, vision understanding, and end-to-end autonomous coding plus long-horizon task execution.
• Benchmark placement per the article: second only to Claude models on TextArena, 4th globally on CodeArena, 2nd on Vision Arena — all leaderboard-style, self-reported positioning rather than independent evaluation.
• Distribution matters as much as the model: same-day availability through third-party agent/coding tools and inference platforms means developers can call it without touching Alibaba's own API.
• Alibaba says Qwen3.8-Max will be open-sourced next week — a first for the Max tier — alongside Qwen3.8-27B, extending an ecosystem it claims spans 400+ released models, 200k+ derivatives, and 1B+ downloads.