有作者分享了使用gpt-5.6-luna等小型人工智能模型的实际体验，指出这类模型在速度、能力和成本方面取得了显著进步，使得消费级与企业级AI应用的落地变得切实可行 1。文章分析认为，过去由于推理成本居高不下，消费级AI公司十分稀缺，而如今市场对“快速、廉价且足够好”的小型模型需求即将迎来爆发，这将深刻改变企业处理日常工作的方式 1。
在具体性能与成本表现上，gpt-5.6-luna模型的运行速度约为100 tps，使用其搜索数千封邮件的API成本仅需几十美分 1。以构建个性化每日新闻网站为例，采用上一代Sonnet级别模型的成本约为1美元，而使用luna模型的平均成本已大幅降至约0.10美元 1。此外，Segment联合创始人Peter（曾为Charm Industrial融资超1亿美元，并为Revoy完成A轮融资）表示，他约95%的工作都属于“大量输出token”类型的响应性任务 1。
Peter, the author of a recent analysis and co-founder of Segment who has raised over $100 million for Charm Industrial and completed Series A funding for Revoy, shared his experience utilizing small AI models such as gpt-5.6-luna 1. He highlighted significant advancements in the speed, capability, and cost-efficiency of these smaller models, which are now making consumer and enterprise AI applications highly viable 1. While the high cost of inference previously caused a scarcity of consumer AI companies, Peter asserts that the demand for "fast, cheap, and good enough" small models is poised to explode, ultimately transforming how enterprises manage their daily operations 1.
To illustrate these advancements, Peter noted that gpt-5.6-luna operates at a speed of approximately 100 tokens per second 1. The cost reductions are substantial compared to previous generations; building a personalized daily news website with an older Sonnet-level model costs around $1, whereas the same application built with luna averages just $0.10 1. Additionally, using gpt-5.6-luna to search through thousands of emails via API costs only a few dozen cents 1. Peter also revealed that roughly 95% of his work involves responsive tasks characterized as "token spewers" 1.
评论
还没有评论，欢迎留下第一条。