On January 22, 2025, ByteDance's Volcano Engine officially launched the Doubao Model 1.5, bringing significant enhancements to its performance capabilities on the Volcano Ark platform. This release underscores ByteDance's commitment to advancing artificial intelligence technology, achieving a globally leading level of overall capability.

The Doubao Model 1.5 comprises several versions, including the Doubao-1.5-pro, which has excelled in multiple authoritative evaluation benchmarks in areas such as knowledge, coding, reasoning, and proficiency in Chinese. This version has outperformed leading industry models like GPT-4o and Claude 3.5 Sonnet. In contrast, the Doubao-1.5-lite model stands out in the lightweight language model category, showcasing performance that rivals that of its predecessor, the Doubao-pro-32k-0828, thus providing users with an improved cost-performance ratio. Additionally, the Doubao-1.5-vision-pro has undergone comprehensive upgrades in multi-modal data synthesis, dynamic resolution, and multi-modal alignment, showing notable advancements in visual reasoning and the understanding of fine-grained information.

The introduction of the Doubao Real-Time Voice Model enables efficient end-to-end voice conversations with low latency and the capacity to interrupt during dialogue, signaling a breakthrough in the field of voice interaction. Volcano Engine is set to launch corresponding API services through the Ark platform in the first half of the year, which will further promote the widespread application of voice technology.

In terms of technical architecture, the Doubao Model 1.5 employs a large-scale sparse MoE architecture, achieving the performance of a dense model with the equivalent of seven times the activation parameters, using a smaller number of active parameters. This innovation significantly exceeds conventional industry efficiency. Furthermore, ByteDance's proprietary server cluster solutions and network card technology help reduce hardware costs, optimize communication efficiency for small packets, and ensure the stability and efficiency of multi-machine distributed inference. Importantly, the training process of Doubao Model 1.5 did not utilize data generated by other models, establishing a completely independent data production system that guarantees the independence and reliability of data sources.

Notably, despite the substantial improvements in performance and features, the price of Doubao Model 1.5 remains unchanged, adhering to the principle of "more features at no extra cost." This strategy aims to promote the accessibility of AI technology, allowing a broader range of enterprises and developers to benefit from these advanced technological achievements.

Key Points

1. ByteDance launched the Doubao Model 1.5 on January 22, 2025.
2. The new model includes several versions with significant performance improvements.
3. The price of Doubao Model 1.5 remains unchanged despite enhanced features.
4. New voice technology allows for real-time interactions with low latency.
5. The training process ensured independence by not using data from other models.