Inception 发布扩散语言模型 Mercury 3，称速度远超 GPT-6 Luna 与 Claude Haiku 5.5
原标题：Inception's Mercury 3 diffusion model claims to beat GPT-6 Luna and Claude Haiku 5.5 at a fraction of the wait
AI 摘要
AI 初创公司 Inception 发布扩散式语言模型 Mercury 3，采用先草拟整体答案再逐步精炼的生成方式，而非逐词输出。Inception 称其在 Nvidia GPU 上每秒生成超 1000 个 token，速度是 GPT-6 Luna 和 Claude Haiku 5.5 的 4 到 8 倍，并在内部 12 项基准中赢下 8 项，但部分长代码和长文本检索测试仍落后。该模型支持 100 万 token 上下文和 OpenAI 兼容 API，主要面向 AI 智能体、智能眼镜和客服等对延迟敏感的场景。
正文节选
Inception's Mercury 3 diffusion model claims to beat GPT-6 Luna and Claude Haiku 5.5 at a fraction of the wait Key Points - AI startup Inception has released Mercury 3, a diffusion-based language model. Instead of writing text word by word, it refines a rough draft of the whole answer step by step. Inception says this makes it much faster than standard models. - In Inception's internal benchmarks, Mercury 3 generates more than 1,000 tokens per second on Nvidia GPUs. It leads rivals like GPT-6 Lu