Inception公司推出了名为Mercury 3的大型语言模型，采用扩散技术架构1。该模型在处理速度上取得重大突破，达到1,068 tokens per second（在NVIDIA GPU上）1，相比现有模型提升4-8倍1。
Mercury 3在多项基准测试中表现突出1。在50个Mind2Web导航任务中的完成率达到50%，超越GPT-6.1-Sol1。Octen search API测试中，模型的中位数生成时间为0.9秒，端到端响应时间为2.2秒1。参数规模方面，Mercury 3比Mercury 1增加了40倍以上1，并支持100万tokens的上下文长度，显著超越Mercury 2.5的260K tokens1。
在定价和应用方面，Mercury 3的输入成本为每百万tokens 0.20美元，输出成本为0.70美元1，发布期间提供50%折扣1。已有早期合作伙伴在浏览器代理、实时教育、搜索代理、客服支持和语音应用等领域展开应用1。
Inception has unveiled Mercury 3, a large language model built on diffusion technology that achieves speeds of 1,068 tokens per second on NVIDIA GPUs 1. The model represents a significant scaling increase, with parameters exceeding those of Mercury 1 by more than 40 times 1.
Mercury 3 supports context lengths of up to 1 million tokens, compared to 260,000 tokens in its predecessor Mercury 2.5 1. The company claims the model outperforms GPT-6 Luna and Claude Haiku 5.5 across multiple benchmarks, while delivering 4 to 8 times faster performance 1. In testing on 50 Mind2Web navigation tasks, Mercury 3 achieved a 50 percent completion rate, surpassing GPT-6.1-Sol 1. Response latency measurements from Inception's search API tests showed a median generation time of 0.9 seconds and an end-to-end response time of 2.2 seconds 1.
Mercury 3 is priced at $0.20 per million input tokens and $0.70 per million output tokens, with a 50 percent discount available during the launch period 1. Early partners are already deploying the model across multiple applications, including browser agents, real-time education, search agents, customer support, and voice applications 1.
评论
还没有评论，欢迎留下第一条。