V4.1-Flash-1017 发布：开源权重、Search 与 Fetch API 各 0.75 美元/千次
据模型卡，V4.1-Flash-1017 提升推理、编程、agent 与视觉跑分，同时缩短执行时间和输出长度；Search 与 Fetch API 对外开放，各 0.75 美元/千次调用，权重在 Hugging Face 和 ModelScope 免费下载。
DeepSeek released V4.1-Flash-1017, an open-weights update to V4.1-Flash, on October 17. Per its model card, it improves reasoning, coding, agentic and visual benchmarks while cutting task execution time and output length The same release opens DeepSeek's Search and Fetch APIs to everyone at $0.75 per 1,000 calls each, powered by an in-house search engine the company says serves hundreds of millions of requests a day The weights are on Hugging Face and ModelScope, free to download. One lab now ships its model weights, its search API, and its benchmark scores in the open on the same day. What does the closed playbook have left?
这条是引用着另一条写的：x.com/deepseek_ai/status/2111375640717033777