ByteDance Seed

ByteDance Seed is ByteDance's advanced artificial intelligence research team, established in 2023 with the goal of pushing the frontiers of artificial general intelligence and developing cutting-edge foundation models. The division operates research laboratories in China, Singapore and the United States.

The team conducts research across multiple disciplines: large language models with pre-training paradigms and efficient architectures, multimodal models integrating vision, language and audio, world models for physical and virtual simulation, AI infrastructure including large-scale training clusters, robotics with reinforcement learning, AI for sciences applied to molecular biology and quantum chemistry, and speech processing for simultaneous interpretation.

Notable developments include the Seed-OSS series of open-source language models optimized for reasoning and multilingualism, Seed-VL for vision-language tasks, BAGEL as unified open-source multimodal model, Seed-Prover for automated theorem proving, Seedream & SeedEdit for image generation and editing, Seedance for video generation, and Seed LiveInterpret for low-latency speech-to-speech interpretation.

Technologies developed by Seed power over 50 ByteDance applications, including Doubao as AI assistant, Coze for agent development, Jimeng for content generation, Lark as enterprise productivity suite, and external services through Volcano Engine.

The team combines foundational research with practical applications, maintaining an open-source approach in several projects while developing technologies that fuel products at global scale.