智能体人工智能(Agent AI)
Search documents
阿里除夕开源“王炸”千问 3.5-Plus ,性能媲美Gemini 3 Pro、Claude 4.5 Opus,百万 Token 8毛钱
AI前线· 2026-02-16 10:45
Core Viewpoint - Alibaba has launched the Qwen 3.5-Plus model, which is positioned to compete directly with Gemini 3 Pro, achieving superior performance in several benchmarks while offering a significantly lower API cost [2][3][4]. Group 1: Model Performance and Cost - Qwen 3.5-Plus has a total parameter scale of 397 billion, with only 17 billion parameters activated at a time, allowing it to outperform models with over a trillion parameters [7][17]. - The API price for Qwen 3.5-Plus is set at 0.8 RMB per million tokens, which is 1/18th of the cost of Gemini 3 Pro [3]. - In various high-difficulty benchmarks, Qwen 3.5-Plus has achieved scores that place it in the "first tier," surpassing models like GPT-5.2 and Claude 4.5 [12]. Group 2: Technical Innovations - The model represents a complete architectural overhaul rather than a simple parameter increase, utilizing a hybrid architecture that combines linear attention mechanisms and sparse mixture of experts (MoE) [6][17]. - Qwen 3.5-Plus has improved inference efficiency, achieving up to 19 times the throughput in long-context scenarios compared to previous models [17][18]. - The model's training has been optimized for multi-modal capabilities, allowing it to handle text, images, and videos effectively, which enhances its application in real-world scenarios [21]. Group 3: Market Reception and Future Implications - The release has generated significant interest in overseas tech communities, with discussions highlighting its UI design capabilities and multi-language support, which now covers 201 languages [25][26]. - The trend towards "Agent AI" is noted, indicating a shift in focus from conversational abilities to actionable intelligence within applications, positioning Qwen 3.5-Plus as a key player in this emerging market [28][29]. - The model's capabilities are expected to extend beyond basic tasks, enabling it to perform complex operations in both mobile and PC environments, thus enhancing overall operational efficiency [21].