[2603.20633] Seed1.8 Model Card: Towards Generalized Real-World Agency

Computer Science > Artificial Intelligence arXiv:2603.20633 (cs) [Submitted on 21 Mar 2026] Title:Seed1.8 Model Card: Towards Generalized Real-World Agency Authors:Bytedance Seed View a PDF of the paper titled Seed1.8 Model Card: Towards Generalized Real-World Agency, by Bytedance Seed View PDF HTML (experimental) Abstract:We present Seed1.8, a foundation model aimed at generalized real-world agency: going beyond single-turn prediction to multi-turn interaction, tool use, and multi-step execution. Seed1.8 keeps strong LLM and vision-language performance while supporting a unified agentic interface-search, code generation and execution, and GUI interaction. For deployment, it offers latency- and cost-aware inference, including configurable thinking modes and optimized visual encoding for images and video. We report evaluations on standard benchmarks and application-aligned workflows spanning foundational skills, multimodal understanding, and agentic behavior. Seed1.8 is released to support further research and development on interactive, real-world use cases. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2603.20633 [cs.AI]   (or arXiv:2603.20633v1 [cs.AI] for this version)   https://doi.org/10.48550/arXiv.2603.20633 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Yujia Qin [view email] [v1] Sat, 21 Mar 2026 04:03:45 UTC (6,596 KB) Full-text links: Access Paper: View a PDF of the paper titled Seed1.8 Model Card: To...