Baidu Open-Sources ERNIE 5.5 Flash Base: MoE With 280B Parameters, 1M-Token Context
Baidu officially open-sourced its latest base model ERNIE 5.5 Flash Base on November 5. Built on a Mixture-of-Experts architecture with 280 billion total parameters and 1…
Baidu officially open-sourced its latest base model ERNIE 5.5 Flash Base on November 5. Built on a Mixture-of-Experts architecture with 280 billion total parameters and 14 billion activated per token, it was pre-trained on 33.5 trillion tokens with a 128K pre-training context and up to 1 million tokens during fine-tuning and inference. Emphasizing efficiency, it introduces Latent-Gated Sparse Attention, Multi-Router Latent MoE and an ECHO training objective, and posted state-of-the-art results among same-size models on knowledge, reasoning, math, code and long-context benchmarks. The model is available on Hugging Face.
Source: Xinhua / Baidu, Nov 5, 2026 | 2026-11-05
延伸阅读：
- Astribot CEO Says Robotics Is Approaching Its 'iPhone 4' Breakthrough Moment
- Xingyun Technology's Compute Orders Top 20.6 Billion Yuan, Plans 200-300M Buyback
- USTC Releases Open-Source RNA Sequence Foundation Model structRFM, Published in Nature
- DeepSeek Closes 105.5 Billion Yuan Round as State Investors Raise Stake
- Tencent Global Digital Ecosystem Conference Spotlights AI Productivity, WorkBuddy Agent
- Chinese Developers Rewriting AI Rules as Open Source Becomes a Global Flag
- MIIT Solicits Comment on Unified Identification-Code Rule for Robot Products
- Economic Daily: AI Development Should Also Be Open and Shared
- Winter Warmers Sell Out: Yiwu Hot-Water Bottles Get Fashion 'Coats' and Become Trendy Toys
- Yiwu Sends 4,435 Buyers to 9th CIIE, Firms Bring Global Goods In
