summary_zh: Baidu 280B 模型可供歐洲創業公司下載,但授權禁止在歐盟、英國及韓國使用於產品開發。
- 模型規格:280B 引數,14B 主動 per token,未經 chat fine‑tune,作為原始權重發布。
AI 伴隨問答
針對這篇報導有疑問?可直接點選快捷問題或向 AI 追問背景脈絡與技術細節:
european startups can download @Baidu_Inc's new 280b model, but its license gives them no right to build products on it
ernie 5.5 flash base came out with open weights and a license that excludes the eu, the uk and south korea. baidu released its previous open models, ernie 4.5, under apache 2.0
meta put a similar limit on its multimodal llama 3.2 models in 2024. baidu has not explained the choice. the eu ai act may be behind it, since the act eases the rules for open models only under a free open-source license
280b parameters with 14b active per token, released as raw base weights without chat tuning. baidu's own tests put it ahead of the deepseek, hy3 and mimo base models on 24 of 30 benchmarks