發布於 2026-08-16 · 赵予安
直接回答
OpenAI GPT-5 Nano 已進入目錄，標價為每百萬 token 0.007 / 0.058，上下文 400,000。該精確 ID 的 CacheSafety 未公布。 這篇指南面向正在比較模型品質、成本、路由策略和正式環境上線風險的台灣產品與平台團隊。
What shipped
OpenAI has a new catalog entry named GPT-5 Nano, with public model id openai/gpt-5-nano and slug openai--gpt-5-nano. The display name is "OpenAI: GPT-5 Nano". The provider is OpenAI. The model has streaming, tool, JSON, vision, and long-context capabilities. Its latency band is 1000 to 3000 ms.
Pricing and context
The listed input price is $0.007 per 1 million tokens. The listed output price is $0.058 per 1 million tokens. The context window is 400,000 tokens. For comparison, the catalog lists GPT-4.1 Nano at $0.014 input and $0.058 output, GPT-4o-mini at $0.022 input and $0.087 output, and GPT-5.4 Nano at $0.029 input and $0.181 output.
Who this is for
The catalog lists vision, long-context, and agent as use cases. That makes this entry relevant for vision workloads, long-document tasks, and agent workflows. Teams looking for a low-cost OpenAI model with a 400,000-token context window have this option in the catalog.
What remains unverified
The release date is unpublished in this catalog snapshot. Independent third-party quality scores are not in these notes. There is no first-party CacheSafety Bench row for this exact model. Safe Hit, Bad Hit, and trap rates are therefore unavailable.