国泰海通｜计算机：豆包大模型1.6发布，性价比进一步提升

Core Viewpoint - ByteDance's recent launch of the Doubao model 1.6 represents a significant upgrade in its AI model family, enhancing both performance and cost-effectiveness, which is expected to directly benefit AI implementation across various industries [2][4]. Group 1: Product Launch and Features - The Doubao model family has been expanded with the release of Doubao model 1.6, Doubao Video Generation Model Seedance 1.0 Pro, and Doubao Real-time Voice Model, establishing itself as a leading model with full modality, full size, and high cost-performance [2][3]. - Doubao model 1.6 includes three models: 1. Doubao-seed-1.6, an all-in-one model supporting 256K context, deep thinking, multi-modal understanding, and graphical interface operations [3]. 2. Doubao-seed-1.6-thinking, a reinforced version focusing on deep thinking capabilities, with improvements in coding, mathematics, and logical reasoning [3]. 3. Doubao-seed-1.6-flash, a high-speed version with low latency and strong visual understanding capabilities [3]. Group 2: Performance Metrics - By May 2025, the daily token usage of Doubao model exceeded 16.4 trillion, marking a 137-fold increase since its initial release in May 2024 [2]. - Doubao model 1.6 has shown significant improvements in reasoning, mathematics, instruction adherence, and agent capabilities, supporting complex agent construction and enhancing AI productivity [2][3]. Group 3: Pricing Strategy - Doubao model 1.6 introduces a novel pricing model based on input length, with a uniform token price regardless of whether deep thinking mode is activated [4]. - For the most commonly used input range of 0-32K, the pricing is set at 0.8 yuan per million tokens for input and 8 yuan per million tokens for output, representing a 63% cost reduction compared to previous models [4]. - A promotional pricing tier has been established for inputs of 32K and outputs within 200 tokens, further lowering costs to 0.8 yuan per million tokens for input and 2 yuan per million tokens for output [4].