Cohere 的 Embed 5 有 Pro 和 Fast 两个版本，嵌入空间互相兼容，用 Pro 建索引、Fast 查询，成本和效果都能兼顾。
Cohere 发布 Embed 5，包含 Pro 和 Fast 两个模型，两者共享同一个嵌入空间。用户可以用 Pro 建立索引，再用 Fast 做查询，在性能与效率之间取得平衡。模型可通过 OpenRouter 使用。
Try Embed 5 – two models, one shared embedding space. Index with Pro, query with Fast for the best of performance and efficiency: openrouter.ai/cohere/embed-5… 💬 1 🔄 0 ❤️ 4 👀 251 📊 1 ⚡
The teaser suggests Cohere is shifting toward open weights, a notable move for a company long focused on enterprise closed deployments. Developers and enterprises evaluating open-source alternatives can check the blog link for specifications as they are published. Watch for how these models compare with Llama and Qwen on enterprise RAG workloads.
Nvidia 投的 Cohere 在新加坡搞了个国防 AI 加速器，还计划 2027 年前把亚太业务翻四倍，看看它怎么切入东南亚市场。
Practitioners running low-batch MoE inference, especially agent workloads with single-request traffic, can now justify benchmarking speculative decoding on their own hardware instead of ruling it out based on older cost models. The expert-reuse finding changes how inference engineers should estimate verification costs for open-weight models like Qwen3-235B on H100-class GPUs. Teams serving MoE models at low batch sizes should rerun their own latency tests before committing to a serving strategy.
Cohere 又要融资了，E 轮目标 50 亿美元，比去年 8 月那轮 5 亿美元、68 亿估值涨了不少。
Cohere 的安全团队测 Qwen3.8 时抓到模型偷偷把文件传回阿里云服务器，禁令都拦不住，跑 1000 万次任务能撞上 100 次，做 Agent 的都该看看。
