使用成本降至三分之一！字节大模型，重磅更新！

Core Viewpoint - The article discusses the significant upgrade of the Doubao large model and the introduction of an innovative "interval pricing" model, marking a "value war" that balances cost and performance [1][3]. Group 1: Model Upgrades - On June 11, ByteDance's Volcano Engine held the Force Original Power Conference, where the Doubao large model family was fully upgraded, including the release of Doubao 1.6 and other new models [3][4]. - Doubao 1.6 is a comprehensive model supporting 256K long context and adaptive reasoning, while Doubao 1.6-thinking enhances deep reasoning capabilities [5][6]. - The Doubao 1.6 series has achieved top global performance in complex reasoning, competitive mathematics, multi-turn dialogue, and instruction adherence [6]. Group 2: Cost Reduction and Pricing Strategy - Doubao 1.6 introduces a groundbreaking "interval pricing" model, reducing overall usage costs to one-third of the previous Doubao 1.5 deep thinking model [7][8]. - The pricing strategy eliminates functional premiums, meaning the token price remains consistent regardless of whether deep reasoning or multimodal capabilities are activated [11]. - The pricing structure is divided into intervals based on input length, with the most used 0-32K input range priced at 0.8 yuan per million tokens for input and 8 yuan for output, significantly lowering costs [12][14]. Group 3: Market Impact and Future Prospects - The pricing strategy aims to address the high cost of enterprise-level agents, which can reach up to $20 per day in token consumption [18]. - Doubao's daily token usage surged from 4 trillion in December last year to 16.4 trillion by May 2025, reflecting a growth rate exceeding 300% [18]. - Volcano Engine holds a 46.4% market share in China's public cloud large model service market, indicating its leading position and potential to accelerate the large-scale application of agents [19].