大模型
www.aibase.com ·
DeepSeek 明确调休日计费规则，中秋国庆迎来十天半价窗口
DeepSeek 于 9 月 19 日发布 API 峰谷时段补充说明，明确调休周末及法定节假日均按低谷价计费。据此，9 月 20 日至 10 月 10 日期间，中秋三天、国庆七天加两个调休周末共十余天均适用低谷价。文中还提及 9 月 10 日发布的 V4.1 Flash 模型，称其为 552B 参数 MoE 架构，输入激活 8B、输出激活 16B，低谷时段输入缓存命中价 0.02 元/百万 Token、输出 4 元/百万 Token，但相关参数与基准表现未提供可验证来源。
AI推荐理由 内容为 DeepSeek 官方对 API 峰谷计费规则的补充说明，明确调休周末按低谷价计费，并给出中秋国庆期间的具体半价窗口，对使用 DeepSeek API 的开发者有直接成本参考价值。但来源为 AIbase 转述，非官方原文，且文中关于 V4.1 Flash 模型参数、定价和基准表现的描述缺乏可验证证据，需谨慎对待。
62
正文
DeepSeek Clarifies Overtime Pay Calculation Rules, Mid-Autumn and National Day Holidays Bring a Ten-Day Half-Price Window
AIbase基地
Published inAI News · 3 min read · Sep 20, 2026
4
On September 19, DeepSeek officially released a supplementary explanation regarding API peak and off-peak hours, clearly stating that weekends with adjusted workdays and national holidays in China are all billed at off-peak rates. This means that developers can enjoy off-peak pricing when calling APIs on September 20 (Sunday with adjusted workday), October 10 (Saturday with adjusted workday), and during the Mid-Autumn Festival and National Day holidays.
According to DeepSeek's previously announced peak and off-peak pricing mechanism, the peak hours on weekdays (9:00-12:00, 14:00-18:00) are twice as expensive as the off-peak hours, while the entire weekend is uniformly billed at the off-peak rate. This supplementary explanation further clarifies the ambiguity of whether "adjusted workdays" are considered weekdays or weekends, explicitly stating that any weekend will be charged at the off-peak rate.
According to this rule, from September 20 to October 10, DeepSeek will experience an extended promotional period: three days of the Mid-Autumn Festival (September 25-27), seven days of National Day (October 1-7), plus two adjusted weekend days, totaling more than ten days all billed at off-peak hours, making it the most concentrated period of "half-price days" of the year.
At the same time, DeepSeek's V4.1 Flash model, released on September 10, further improves the cost-effectiveness. This 552B parameter MoE model uses an asymmetric architecture, activating only 8B parameters for input and 16B parameters for output, yet outperforms its flagship V4Pro model in multiple benchmark tests, while being priced at the Flash tier. During off-peak hours, the input cache hit rate for V4.1 Flash is as low as 0.02 yuan per million Tokens, and the output is only 4 yuan per million Tokens.
DeepSeekAPI Peak-Valley PricingOvertime WorkLow Valley Price
This article is from AIbase Daily
Scan to view
Welcome to the [AI Daily] column! This is your daily guide to exploring the world of artificial intelligence. Every day, we present you with hot topics in the AI field, focusing on developers, helping you understand technical trends, and learning about innovative AI product applications.
—— Created by the AIbase Daily Team
© Copyright AIbase Base 2024, Click to View Source -