Qwen 3.8 Max 0902

Qwen3.8-Max-0902 (alias qwen3.8-max-2026-09-02) is an upgraded snapshot of Alibaba's proprietary qwen3.8-max model, served through QwenCloud. According to the vendor, it improves coding on larger engineering-scale projects and long-horizon autonomous development, and shows better multi-tool orchestration and end-to-end task delivery in collaborative agent settings.\n\nNative vision understanding is also refined, covering chart reasoning, document parsing and multimodal perception. The model keeps the 1M-token context window, a thinking mode, and the existing tool ecosystem. The listed limits are 991K max input, 131K max output, and 262K max reasoning tokens. A YouTube review described it as the same architecture as the previous version (reported as about 2.4T parameters), post-trained further on coding and co-work; this parameter figure is not confirmed on the vendor page.\n\nIt is accessed through an OpenAI-compatible API and supports prefix completion, function calling, structured outputs, batches, web search, caching and fine-tuning. It can also be used through the QwenCloud Token Plan with tools such as Qwen Code, Cline, Claude Code and Cursor. It is not open source and no weights are published.