博客
生成更真实，百轮编辑更稳定｜Seedream 5.5 发布
非编辑区域像素级保持，真实感与小字渲染显著提升
SeedRealtime 音视频全双工大模型发布：走向全模态自然交互
作为原生音视频全双工大模型，联合理解声音、画面与时序信息，带来边看、边听、边说的自然交互体验
一镜成片，随心参考｜Seedance 2.5 正式发布
30 秒长叙事，多模态参考能力、编辑能力全面升级
从“会说”走向“会创作”｜ Seed Audio 1.0 音频创作模型发布
统一框架下联合建模人声、音效和环境声等音频要素
EdgeBench：衡量真实世界环境学习，发现新 Scaling Law
Agent 在环境学习中表现出明显的 log-sigmoid 规律
精选论文
2026.10.25Recycler: Cost-Free Latent Reasoning via Hidden State Recycling2026.09.28Periodic Weak Spots: Phase Sensitivity from Chunked KV-Cache Compression2026.07.06EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments2026.06.30Seed2.0 Model Card: Towards Intelligence Frontier for Real-World Complexity2026.05.29Task-Focused Memorization for Multimodal Agents2026.05.16Charon: A Unified and Fine-Grained Simulator for Large-Scale LLM Training and Inference2026.04.26MegaScale-Omni: A Hyper-Scale, Workload-Resilient System for MultiModal LLM Training in Production2026.04.22Seed3D 2.0: Advancing High-Fidelity Simulation-Ready 3D Content Generation2026.04.15Seedance 2.0: Advancing Video Generation for World Complexity2026.04.11Protenix-v2: Broadening the Reach of Structure Prediction and Biomolecular Design