IT之家 9 月 30 日消息,在今天召开的 2026 年 OpenAI 开发者日活动日中,官方推出 Ultrafast 全新服务层级,在 Codex 中可给出最高 8 倍词元(Token)生成速度,在 API 调用时最高可以达成 6 倍生成速度。
API 调用费用维度,IT之家查询官方信息,附上售价如下:
| 模型 | 售价类型 | 短上下文(输入< 272K Token) | 长上下文(输入> 272K Token) |
|---|---|---|---|
| gpt-6-astra | 输入(Input) | $60.00 / 1M tokens | $120.00 / 1M tokens |
| gpt-6-astra | 缓存输入(Cached input) | $6.00 / 1M tokens | $12.00 / 1M tokens |
| gpt-6-astra | 缓存写入(Cache writes) | $75.00 / 1M tokens | $150.00 / 1M tokens |
| gpt-6-astra | 输出(Output) | $300.00 / 1M tokens | $450.00 / 1M tokens |
OpenAI 官宣 GPT‑6 Astra Ultrafast 在 Codex 中可以达成每秒 300 个词元,当下已开放 API 调用,Pro 500 订阅和企业使用者可以在 ChatGPT Work 和 Codex 中感受,官方表示后续将会跟进推出 GPT-6.1 Sol Ultrafast。
Ultrafast 的价值在于“更强模型”与“实时响应”兼得,过去要达成接近实时的速度,往往需要换用更小或更专用模型;而 Ultrafast 试图在保持 GPT‑5.6 Sol 能力的前提下,把“每秒有用工作量”做大,从而更合适事件响应、实时客服、交易 / 市面快讯生成、交互式编码等情形。
本文来源:https://www.ithome.com/1/008/524.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
IT Home reported on September 30 that at the 2026 OpenAI Developer Day held today, the company launched the new Ultrafast service tier, which delivers up to 8x token generation speed in Codex and up to 6x generation speed in API calls.
In terms of API call pricing, IT Home checked official information and lists the prices as follows:
OpenAI officially announced that GPT-6 Astra Ultrafast can achieve 300 tokens per second in Codex, and API access is now open. Pro 500 subscription and enterprise users can experience it in ChatGPT Work and Codex. The official stated that GPT-6.1 Sol Ultrafast will be rolled out subsequently.
The value of Ultrafast lies in combining a "stronger model" with "real-time response." In the past, achieving near-real-time speed often required switching to smaller or more specialized models; Ultrafast, however, aims to increase "useful work per second" while maintaining GPT-5.6 Sol's capabilities, making it more suitable for event response, real-time customer service, trading/market news flash generation, interactive coding, and similar scenarios.


