DeepSeek Raised Prices Up to 1,114%. What You'll Actually Pay Is a Different Number.

Published: August 17, 2026 Impact: High. Affects every developer and business currently paying for DeepSeek's API.

What Changed

DeepSeek's API moved to peak and off-peak pricing this week, widely reported to take effect at 16:00 UTC on August 16, 2026. Tokens are the chunks of text AI companies charge by, about three-quarters of a word each, split into input (what you send) and output (what the model sends back). Both now cost different amounts by the hour.

Every outlet is quoting a 1,114% jump. That figure is real. It's also the smallest charge on the bill, not the number that will actually move your invoice. The real driver is output pricing, up 355% to 371% at peak. A coding session that ran about $2 in output charges last week can run roughly $9 today at peak, about $4.50 off-peak.

The Real Rate Card

I checked api-docs.deepseek.com/quick_start/pricing/ directly. DeepSeek states it plainly: "Off-peak rates are half of the peak rates. Peak hours are 01:00-04:00 and 06:00-10:00 UTC (all other hours are off-peak)." Per million tokens, old rate versus new off-peak/peak range:

• V4-Pro output: $0.87 → $1.98 / $3.96 (+127% / +355%)
• V4-Flash output: $0.28 → $0.66 / $1.32 (+136% / +371%)
• V4-Pro cache-hit input, repeated text the model already processed: $0.003625 → $0.022 / $0.044, the +1,114% figure everyone is repeating, still under 5 cents at peak

Cache-hit pricing starts from a fraction of a cent, so even an elevenfold jump lands small in absolute terms. Output carries no such discount, and moved the most in real dollars.

"Off-Peak Is Cheap" Isn't the Whole Story

DeepSeek's peak window, 01:00-10:00 UTC with a two-hour gap, is mid-morning into midday across China and South Asia, and the middle of the night across the Americas and most of Europe. A developer in Austin or Berlin dodges peak pricing just by working normal hours. A developer in Shanghai or Bangalore, working those same hours, is far more likely to land inside it.

Three Days, Versus Ninety

Coverage of this change surfaced August 13-14 (Yahoo Finance, Fortune, Engadget, TechNode), days before the Aug 16 effective date those reports cite. DeepSeek's own pricing page carries no timestamp; the date comes from that coverage, not DeepSeek directly.

Compare that to April 24, 2026, when DeepSeek's changelog gave the older deepseek-chat and deepseek-reasoner API names a 90-day retirement notice. Three days now, ninety days then, for changes of comparable weight. Rising compute costs are a plausible, ordinary reason to reprice, not unique to DeepSeek. The issue isn't that prices moved. It's that notice shrank by a factor of thirty while the loudest number is the one least likely to hit anyone's actual bill.

What This Resolves

I flagged this coming on August 6, when DeepSeek posted a vague warning of a "significant" increase with no figures or date (BNA #20). This is that number. It also updates BNA #18: DeepSeek still beats OpenAI's cut Luna rate on output, but the gap narrowed sharply at peak hours. OneHuman's compare page already reflects the new ranges.

What to Do Next

Budgeting for DeepSeek API usage: plan around peak numbers, not the old flat rate.

Comparing DeepSeek to competitors: use output pricing, not the cache-hit percentage making headlines.

Working standard East or South Asia hours: you're least able to dodge peak pricing by rescheduling.

Sources

• DeepSeek API pricing page, checked August 17, 2026
• DeepSeek API updates changelog, April 24, 2026 entry
• Yahoo Finance, Fortune, Engadget, TechNode, DeepSeek pricing coverage, August 13-14, 2026
• OneHuman BNA #20, "DeepSeek Warned Prices Are Going Up. It Won't Say How Much or When." (August 12, 2026)
• OneHuman BNA #18, "OpenAI Cut Luna's API Price 80%. DeepSeek Is Still Cheaper." (August 1, 2026)

Verified by OneHuman · August 17, 2026

Independent. AI-assisted. Human-verified. No ads. No affiliates. No investors.