更新时间:2026-08-27 | 核验来源:DeepSeek 官方 API 定价页 + 官方公告 + 第三方对比
🔥 8月定价大事回顾
- 8月17日:DeepSeek 启用峰谷定价,高峰时段价格是低谷时段的 2倍
- 8月23日:进一步优化——周末(周六日)全天按低谷价格计费,取消峰谷区分
📊 新旧价格对比(缓存命中后)
| 模型 |
原价格(元/百万token) |
峰值价格(元/百万token) |
低谷价格(元/百万token) |
周末价格 |
| V4-Pro |
输入 0.025 / 输出 0.1 |
输入 0.30 / 输出 0.90 |
输入 0.15 / 输出 0.45 |
同低谷 |
| V4-Flash |
输入 0.02 / 输出 0.06 |
输入 0.24 / 输出 0.72 |
输入 0.12 / 输出 0.36 |
同低谷 |
| V4-Coder |
输入 0.02 / 输出 0.06 |
输入 0.24 / 输出 0.72 |
输入 0.12 / 输出 0.36 |
同低谷 |
💡 涨幅最高达 1100%(V4-Pro 峰值 vs 原价)
⏰ 峰谷时段划分(工作日)
- 高峰时段:北京时间 9:00-12:00、14:00-18:00
- 低谷时段:12:00-14:00、18:00-次日 9:00
- 周末(周六日):全天按低谷价格计费,不分峰谷
🛠️ 开发者省钱策略
1. 开启 Prompt Caching(最有效)
from openai import OpenAI
client = OpenAI(
api_key="YOUR_API_KEY",
base_url="https://api.deepseek.com"
)
# 长 System Prompt 自动缓存
response = client.chat.completions.create(
model="deepseek-chat",
messages=[
{"role": "system", "content": "你是一个资深 Python 工程师,精通算法优化和系统设计..."},
{"role": "user", "content": "帮我写一个快速排序的优化版本"}
],
cache_control="enable" # 启用缓存
)
缓存命中后,输入 Token 价格降低 80-90%,大幅抵消峰谷差价
2. 错峰调用
- 工作日低谷:12:00-14:00、18:00-次日 9:00
- 周末全天:无峰谷,统一按低谷价
- 批量任务:安排在周末或深夜执行
3. 模型选择建议
| 场景 |
推荐模型 |
理由 |
| 代码生成/Agent |
V4-Coder |
编程专精,性价比最高 |
| 日常对话/RAG |
V4-Flash |
速度快,成本低 |
| 复杂推理/多步规划 |
V4-Pro |
性能最强,但价格较高 |
⚠️ 注意事项
- 缓存未命中时按全价计费:确保 System Prompt 足够长(>1K tokens)才能触发缓存
- 周末价格适用于所有天:周六日统一谷价,但节假日需看官方公告
- 个人用户免费:DeepSeek 网页版/App 聊天功能仍完全免费,仅 API 收费
- 调价不可逆:8月17日后新价生效,历史订单按旧价结算
🔗 官方资源
- DeepSeek API 文档:https://platform.deepseek.com
- 定价页:https://platform.deepseek.com/api-docs/pricing/
- 官方公告:https://platform.deepseek.com/api-docs/changelog/
下一步:需要更多免费 API 选择?查看 智谱 GLM-4.7-Flash 永久免费 API 或 MiniMax 8月限时免费活动。
Last Updated: 2026-08-27 | Verified Sources: DeepSeek Official API Pricing Page + Official Announcements + Third-party Comparison
🔥 August 2026 Pricing Changes
- Aug 17: DeepSeek launched peak/valley pricing — peak hours are 2x valley rates
- Aug 23: Further optimization — weekends (Sat/Sun) now all-day valley pricing, no peak/valley distinction
📊 Price Comparison (After Cache Hit)
| Model |
Old Price (CNY/1M tokens) |
Peak Price (CNY/1M tokens) |
Valley Price (CNY/1M tokens) |
Weekend Price |
| V4-Pro |
Input 0.025 / Output 0.1 |
Input 0.30 / Output 0.90 |
Input 0.15 / Output 0.45 |
Same as valley |
| V4-Flash |
Input 0.02 / Output 0.06 |
Input 0.24 / Output 0.72 |
Input 0.12 / Output 0.36 |
Same as valley |
| V4-Coder |
Input 0.02 / Output 0.06 |
Input 0.24 / Output 0.72 |
Input 0.12 / Output 0.36 |
Same as valley |
💡 Increase up to 1100% (V4-Pro peak vs old price)
⏰ Peak/Valley Hours (Weekdays)
- Peak Hours: 9:00-12:00, 14:00-18:00 (Beijing Time)
- Valley Hours: 12:00-14:00, 18:00-9:00 next day
- Weekends (Sat/Sun): All-day valley pricing, no peak/valley distinction
🛠️ Developer Savings Strategies
1. Enable Prompt Caching (Most Effective)
from openai import OpenAI
client = OpenAI(
api_key="YOUR_API_KEY",
base_url="https://api.deepseek.com"
)
# Long System Prompt auto-caches
response = client.chat.completions.create(
model="deepseek-chat",
messages=[
{"role": "system", "content": "You are a senior Python engineer, expert in algorithm optimization and system design..."},
{"role": "user", "content": "Help me write an optimized quicksort"}
],
cache_control="enable" # Enable caching
)
After cache hit, input token cost drops 80-90%, significantly offsetting peak/valley difference
2. Off-Peak Calling
- Weekday Valley: 12:00-14:00, 18:00-9:00 next day
- Weekend All-Day: No peak/valley, unified valley rate
- Batch Tasks: Schedule for weekends or late night
3. Model Selection Guide
| Scenario |
Recommended Model |
Reason |
| Code Generation/Agent |
V4-Coder |
Coding specialist, best cost-performance |
| Daily Chat/RAG |
V4-Flash |
Fast, low cost |
| Complex Reasoning/Multi-step |
V4-Pro |
Best performance, but higher price |
⚠️ Important Notes
- Full price when cache miss: Ensure System Prompt is long enough (>1K tokens) to trigger cache
- Weekend pricing applies to all days: Sat/Sun unified valley rate, holidays subject to official announcement
- Free for personal users: DeepSeek web/app chat still completely free, only API charges
- Pricing is non-reversible: New prices effective after Aug 17, historical orders settled at old rates
🔗 Official Resources
Next Step: Need more free API options? Check out Zhipu GLM-4.7-Flash Free Forever API or MiniMax August Limited-Time Free Event.