AI 服務暫時不可用，以下為來源正文，待恢復後補全翻譯。
AI models Qwen3.8-Max: Alibaba's new frontier reasoning model Qwen3.8-Max is Alibaba's new open-weight frontier model, featuring a 1M token context window and multimodal input. Authors Albert Lee Last updated August 12, 2026 Share TL;DR Qwen3.8-Max is Alibaba's latest frontier reasoning model. It's the first open-weight model in the Qwen3-Max series, accepts multimodal input, has a 1M token context window, and scores in the top 10 across 185 models on Artificial Analysis' Intelligence Index. Qwen3.8-Max is available now on Baseten Dedicated Inference. What's new in Qwen3.8-Max Qwen3.8-Max represents a significant leap over its predecessor, Qwen3.7-Max (released May 2026). The Artificial Analysis Intelligence Index has climbed from 47 to 58. Notably, this is the first model in the Qwen3-Max series to release with open weights, under the Qwen 3.8-Max license. Here's how the Qwen3-Max series has evolved since it was first released in September 2025. Features and benchmark performance Qwen3.8-Max is a Mixture-of-Experts model with 2.4 trillion total parameters and 95B active parameters per token. Here are a few of its other key features: Configurable reasoning level Multimodal input 1M token context window Function calling and structured outputs Qwen3.8-Max scores 58 on the Artificial Analysis Intelligence Index (i.e., general intelligence), ranking #9 across all 185 benchmarked models. Qwen3.8-Max also ranks in the top 10% across these key quality benchmarks: Ideal use cases Knowledge work and research: Qwen3.8-Max excels at reasoning across large volumes of professional and scientific literature. With its 1M context window and top performance on GDPval-AA v2, AA-LCR, and GPQA Diamond, it provides reliable, high-level reasoning across extensive document sets. Business process automation: Beyond standard question-and-answer tasks, Qwen3.8-Max’s function calling and structured outputs features enable it to execute business workflows utilizing enterprise applications. Strong scores on AutomationBench-AA, APEX-Agents-AA, and τ³‑Banking validate its ability to complete multi-step, cross-application tasks reliably. Scientific and technical problem-solving: Qwen3.8-Max handles technical challenges, from writing executable Python code to solving graduate-level physics problems. Its proficiency is validated by high marks on the SciCode, CritPt, and GPQA Diamond benchmarks. Agentic AI assistants: Create agents that handle varied inputs and multi-step tasks. With multimodal capabilities, configurable reasoning, and high scores on IFBench and MMMU-Pro, it follows intricate instructions across text and visual media with high precision. Available on Baseten Dedicated Inference Qwen3.8-Max is available now on Baseten Dedicated Inference. If you are interested in deploying it, view it in our model library or contact our team to get started. Subscribe to our newsletter Stay up to date on model performance, inference infrastructure, and more.